CREATE INDEX CONCURRENTLY on partitioned index

From 05cd4feaa5a1d77a1ae8e3a167a68b2a21af1fc6 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH v11 1/9] Allow CREATE INDEX CONCURRENTLY on partitioned table

Note, this effectively reverts 050098b14, so take care to not reintroduce the
bug it fixed.

XXX: does pgstat_progress_update_param() break other commands progress ?
---
 doc/src/sgml/ref/create_index.sgml     |   9 --
 src/backend/commands/indexcmds.c       | 141 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 ++++++++++-
 src/test/regress/sql/indexing.sql      |  18 +++-
 4 files changed, 172 insertions(+), 56 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 29dee5689e..bd4431a3ce 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -661,15 +661,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index ca24620fd0..76219381c1 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -671,17 +672,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1119,6 +1109,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1171,6 +1166,14 @@ DefineIndex(Oid relationId,
 		 */
 		if (!stmt->relation || stmt->relation->inh)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			PartitionDesc partdesc = RelationGetPartitionDesc(rel);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
@@ -1178,12 +1181,15 @@ DefineIndex(Oid relationId,
 			TupleDesc	parentDesc;
 			Oid		   *opfamOids;
 
+			// If concurrent, maybe this should be done after excluding indexes which already exist ?
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1226,10 +1232,12 @@ DefineIndex(Oid relationId,
 					continue;
 				}
 
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				childidxs = RelationGetIndexList(childrel);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1300,10 +1308,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1355,10 +1367,18 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
 
@@ -1368,51 +1388,42 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
-	if (!concurrent)
+	table_close(rel, NoLock);
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
 		return address;
 	}
 
-	/* save lockrelid and locktag for below, then close rel */
+	/* save lockrelid and locktag for below */
 	heaprelid = rel->rd_lockInfo.lockRelId;
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
-	table_close(rel, NoLock);
 
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
@@ -1606,6 +1617,56 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexMultipleInternal(partitions, REINDEXOPT_CONCURRENTLY);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index c93f4470c9..f04abc6897 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c..3d4b6e9bc9 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.17.0

From ba40cea72e3f47e5e4d187316766083db5963fc8 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 16:23:02 -0500
Subject: [PATCH v11 2/9] Add SKIPVALID flag for more integration

---
 src/backend/commands/indexcmds.c | 54 +++++++++++---------------------
 src/include/nodes/parsenodes.h   |  7 +++--
 2 files changed, 22 insertions(+), 39 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 76219381c1..ffc166206b 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1617,53 +1617,30 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
-
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, childs)
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
-
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
-	}
-
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 */
-	ReindexMultipleInternal(partitions, REINDEXOPT_CONCURRENTLY);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId,
+			REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -2947,6 +2924,11 @@ ReindexPartitions(Oid relid, int options, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip invalid indexes, if requested */
+		if ((options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h
index d1f9ef29ca..1373b57ae7 100644
--- a/src/include/nodes/parsenodes.h
+++ b/src/include/nodes/parsenodes.h
@@ -3347,10 +3347,11 @@ typedef struct ConstraintsSetStmt
  */
 
 /* Reindex options */
-#define REINDEXOPT_VERBOSE (1 << 0) /* print progress info */
+#define REINDEXOPT_VERBOSE	(1 << 0) /* print progress info */
 #define REINDEXOPT_REPORT_PROGRESS (1 << 1) /* report pgstat progress */
-#define REINDEXOPT_MISSING_OK (1 << 2)	/* skip missing relations */
-#define REINDEXOPT_CONCURRENTLY (1 << 3)	/* concurrent mode */
+#define REINDEXOPT_MISSING_OK	(1 << 2)	/* skip missing relations */
+#define REINDEXOPT_CONCURRENTLY	(1 << 3)	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	(1 << 4)	/* skip valid indexes */
 
 typedef enum ReindexObjectType
 {
-- 
2.17.0

From b1ff160a715a7e87c93bf4f7179086e406f5dad1 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 23:52:31 -0500
Subject: [PATCH v11 3/9] ReindexPartitions() to set indisvalid..

Something like this should probably have been included in
a6642b3ae060976b42830b7dc8f29ec190ab05e4

See also 71a05b223, which mentioned the absence of any way to validate an
index.
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index ffc166206b..c6a75e887f 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1628,8 +1628,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1640,9 +1638,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 */
 	ReindexPartitions(indexRelationId,
 			REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -2944,6 +2939,24 @@ ReindexPartitions(Oid relid, int options, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, options);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.17.0

From 2840a6d355961ea6fdd29c2851b9c333c17c849f Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH v12 1/5] Allow CREATE INDEX CONCURRENTLY on partitioned table

Note, this effectively reverts 050098b14, so take care to not reintroduce the
bug it fixed.

XXX: does pgstat_progress_update_param() break other commands progress ?
---
 doc/src/sgml/ref/create_index.sgml     |   9 --
 src/backend/commands/indexcmds.c       | 142 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 ++++++++++-
 src/test/regress/sql/indexing.sql      |  18 +++-
 4 files changed, 173 insertions(+), 56 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index a5271a9f8f..6869a18968 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -686,15 +686,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index f9f3ff3b62..c513e8a6bd 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -680,17 +681,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1128,6 +1118,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1183,18 +1178,29 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
 			bool		invalidate_parent = false;
 			TupleDesc	parentDesc;
 			Oid		   *opfamOids;
 
+			// If concurrent, maybe this should be done after excluding indexes which already exist ?
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1237,10 +1243,12 @@ DefineIndex(Oid relationId,
 					continue;
 				}
 
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				childidxs = RelationGetIndexList(childrel);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1311,10 +1319,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1366,10 +1378,18 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
 
@@ -1379,51 +1399,42 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
-	if (!concurrent)
+	table_close(rel, NoLock);
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
 		return address;
 	}
 
-	/* save lockrelid and locktag for below, then close rel */
+	/* save lockrelid and locktag for below */
 	heaprelid = rel->rd_lockInfo.lockRelId;
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
-	table_close(rel, NoLock);
 
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
@@ -1617,6 +1628,57 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+	ReindexParams params = { .options = REINDEXOPT_CONCURRENTLY };
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexMultipleInternal(partitions, &params);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index c93f4470c9..f04abc6897 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c..3d4b6e9bc9 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.17.0

From 2ce39870407fa8bb486df75f51bc9f20dd626045 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 16:23:02 -0500
Subject: [PATCH v12 2/5] Add SKIPVALID flag for more integration

---
 src/backend/commands/indexcmds.c | 52 +++++++++++---------------------
 src/include/catalog/index.h      |  1 +
 2 files changed, 18 insertions(+), 35 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index c513e8a6bd..89c2f62eda 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1628,54 +1628,31 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
 	ReindexParams params = { .options = REINDEXOPT_CONCURRENTLY };
 
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, childs)
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
-
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
-	}
-
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 */
-	ReindexMultipleInternal(partitions, &params);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId, &params, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -3037,6 +3014,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip invalid indexes, if requested */
+		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index 266f8950dc..6673122ec2 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -40,6 +40,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.17.0

From 15643b52a721f4c43d00296cd87f152543b018f7 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 23:52:31 -0500
Subject: [PATCH v12 3/5] ReindexPartitions() to set indisvalid..

Something like this should probably have been included in
a6642b3ae060976b42830b7dc8f29ec190ab05e4

See also 71a05b223, which mentioned the absence of any way to validate an
index.
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 89c2f62eda..739bd14001 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1641,8 +1641,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1652,9 +1650,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * this commits and then starts a new transaction immediately.
 	 */
 	ReindexPartitions(indexRelationId, &params, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -3034,6 +3029,24 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, params);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.17.0

From 66c0421963c7f98b004c21ab1b29dc2cd2713d2e Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sun, 1 Nov 2020 12:25:15 -0600
Subject: [PATCH v12 4/5] Refactor to allow reindexing all index partitions at
 once

---
 src/backend/commands/indexcmds.c           | 262 ++++++++++++++-------
 src/test/regress/expected/create_index.out |   4 +-
 2 files changed, 185 insertions(+), 81 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 739bd14001..08d44a1999 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -102,6 +102,8 @@ static void ReindexMultipleInternal(List *relids,
 									ReindexParams *params);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
+static List *ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
+										int options, MemoryContext private_context);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
 
@@ -124,6 +126,15 @@ typedef struct ReindexErrorInfo
 	char		relkind;
 } ReindexErrorInfo;
 
+/* Argument to ReindexIndexConcurrently takes a List* of these */
+typedef struct ReindexIndexInfo
+{
+	Oid			indexId;
+	Oid			tableId;
+	Oid			amId;
+	bool		safe;		/* for set_indexsafe_procflags */
+} ReindexIndexInfo;
+
 /*
  * CheckIndexCompatible
  *		Determine whether an existing index definition is compatible with a
@@ -1635,7 +1646,9 @@ DefineIndex(Oid relationId,
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ReindexParams params = { .options = REINDEXOPT_CONCURRENTLY };
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID,
+	};
 
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
@@ -2607,7 +2620,15 @@ ReindexIndex(RangeVar *indexRelation, ReindexParams *params, bool isTopLevel)
 		ReindexPartitions(indOid, params, isTopLevel);
 	else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
 			 persistence != RELPERSISTENCE_TEMP)
-		ReindexRelationConcurrently(indOid, params);
+	{
+		ReindexIndexInfo idxinfo = {
+			.indexId = indOid,
+			/* other fields set later */
+		};
+		ReindexIndexesConcurrently(list_make1(&idxinfo),
+				list_make1_oid(IndexGetRelation(indOid, false)),
+				params->options, CurrentMemoryContext);
+	}
 	else
 	{
 		ReindexParams newparams = *params;
@@ -2938,20 +2959,69 @@ reindex_error_callback(void *arg)
 				   errinfo->relnamespace, errinfo->relname);
 }
 
+
+/*
+ * Given a list of index oids, return a list of leaf partitions by removing
+ * any intermediate parents.  heaprels is populated with the corresponding
+ * tables.
+ */
+static List *
+leaf_partitions(List *inhoids, int options, List **heaprels)
+{
+	List		*partitions = NIL;
+	ListCell	*lc;
+
+	foreach(lc, inhoids)
+	{
+		Oid			partoid = lfirst_oid(lc);
+		Oid			tableoid;
+		Relation	table;
+		char		partkind = get_rel_relkind(partoid);
+
+		/*
+		 * This discards partitioned indexes and foreign tables.
+		 */
+		if (!RELKIND_HAS_STORAGE(partkind))
+			continue;
+
+		Assert(partkind == RELKIND_INDEX);
+
+		/* Skip invalid indexes, if requested */
+		if ((options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
+		/* (try to) Open the table, with lock */
+		tableoid = IndexGetRelation(partoid, false);
+		table = table_open(tableoid, ShareLock);
+		table_close(table, NoLock);
+
+		/* Save partition OID in current MemoryContext */
+		partitions = lappend_oid(partitions, partoid);
+		*heaprels = lappend_oid(*heaprels, tableoid);
+	}
+
+	return partitions;
+}
+
+
 /*
  * ReindexPartitions
  *
  * Reindex a set of partitions, per the partitioned index or table given
  * by the caller.
+ * XXX: should be further refactored with logic from ReindexRelationConcurrently
  */
 static void
 ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 {
-	List	   *partitions = NIL;
+	List	   *partitions = NIL,
+			*heaprels = NIL;
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
 	MemoryContext reindex_context;
+	MemoryContext old_context;
 	List	   *inhoids;
 	ListCell   *lc;
 	ErrorContextCallback errcallback;
@@ -2996,38 +3066,58 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 * The list of relations to reindex are the physical partitions of the
 	 * tree so discard any partitioned table or index.
 	 */
-	foreach(lc, inhoids)
-	{
-		Oid			partoid = lfirst_oid(lc);
-		char		partkind = get_rel_relkind(partoid);
-		MemoryContext old_context;
 
-		/*
-		 * This discards partitioned tables, partitioned indexes and foreign
-		 * tables.
-		 */
-		if (!RELKIND_HAS_STORAGE(partkind))
-			continue;
-
-		/* Skip invalid indexes, if requested */
-		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
-				get_index_isvalid(partoid))
-			continue;
+	if (relkind == RELKIND_PARTITIONED_INDEX)
+	{
+		old_context = MemoryContextSwitchTo(reindex_context);
+		partitions = leaf_partitions(inhoids, params->options, &heaprels);
+		MemoryContextSwitchTo(old_context);
+	} else {
+		/* Loop over parent tables */
+		foreach(lc, inhoids)
+		{
+			Oid		partoid = lfirst_oid(lc);
+			Relation parttable;
+			List	*partindexes;
+
+			parttable = table_open(partoid, ShareLock);
+			old_context = MemoryContextSwitchTo(reindex_context);
+			partindexes = RelationGetIndexList(parttable);
+			partindexes = leaf_partitions(partindexes, params->options, &heaprels);
+			partitions = list_concat(partitions, partindexes);
+
+			MemoryContextSwitchTo(old_context);
+			table_close(parttable, ShareLock);
+		}
+	}
 
-		Assert(partkind == RELKIND_INDEX ||
-			   partkind == RELKIND_RELATION);
+	if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
+		relkind == RELKIND_PARTITIONED_INDEX &&
+		get_rel_persistence(relid) != RELPERSISTENCE_TEMP)
+	{
+		List			   *idxinfos = NIL;
+		ReindexIndexInfo	*idxinfo;
 
-		/* Save partition OID */
 		old_context = MemoryContextSwitchTo(reindex_context);
-		partitions = lappend_oid(partitions, partoid);
+		foreach (lc, partitions)
+		{
+			Oid partoid = lfirst_oid(lc);
+			idxinfo = palloc(sizeof(ReindexIndexInfo));
+			idxinfo->indexId = partoid;
+			/* other fields set later */
+			idxinfos = lappend(idxinfos, idxinfo);
+		}
 		MemoryContextSwitchTo(old_context);
-	}
 
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 */
-	ReindexMultipleInternal(partitions, params);
+		/* Process all indexes in a single loop */
+		ReindexIndexesConcurrently(idxinfos, heaprels, params->options, reindex_context);
+	} else {
+		/*
+		 * Process each partition listed in a separate transaction.  Note that
+		 * this commits and then starts a new transaction immediately.
+		 */
+		ReindexMultipleInternal(partitions, params);
+	}
 
 	/*
 	 * If indexes exist on all of the partitioned table's children, and we
@@ -3172,18 +3262,9 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 static bool
 ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 {
-	typedef struct ReindexIndexInfo
-	{
-		Oid			indexId;
-		Oid			tableId;
-		Oid			amId;
-		bool		safe;		/* for set_indexsafe_procflags */
-	} ReindexIndexInfo;
 	List	   *heapRelationIds = NIL;
 	List	   *indexIds = NIL;
 	List	   *newIndexIds = NIL;
-	List	   *relationLocks = NIL;
-	List	   *lockTags = NIL;
 	ListCell   *lc,
 			   *lc2;
 	MemoryContext private_context;
@@ -3192,13 +3273,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 	char	   *relationName = NULL;
 	char	   *relationNamespace = NULL;
 	PGRUsage	ru0;
-	const int	progress_index[] = {
-		PROGRESS_CREATEIDX_COMMAND,
-		PROGRESS_CREATEIDX_PHASE,
-		PROGRESS_CREATEIDX_INDEX_OID,
-		PROGRESS_CREATEIDX_ACCESS_METHOD_OID
-	};
-	int64		progress_vals[4];
 
 	/*
 	 * Create a memory context that will survive forced transaction commits we
@@ -3449,6 +3523,69 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 	Assert(heapRelationIds != NIL);
 
+	/* Do the work */
+	newIndexIds = ReindexIndexesConcurrently(indexIds, heapRelationIds, params->options, private_context);
+
+	/* Log what we did */
+	if ((params->options & REINDEXOPT_VERBOSE) != 0)
+	{
+		if (relkind == RELKIND_INDEX)
+			ereport(INFO,
+					(errmsg("index \"%s.%s\" was reindexed",
+							relationNamespace, relationName),
+					 errdetail("%s.",
+							   pg_rusage_show(&ru0))));
+		else
+		{
+			foreach(lc, newIndexIds)
+			{
+				Oid			indOid = lfirst_oid(lc);
+				ereport(INFO,
+						(errmsg("index \"%s.%s\" was reindexed",
+								get_namespace_name(get_rel_namespace(indOid)),
+								get_rel_name(indOid))));
+				/* Don't show rusage here, since it's not per index. */
+			}
+
+			ereport(INFO,
+					(errmsg("table \"%s.%s\" was reindexed",
+							relationNamespace, relationName),
+					 errdetail("%s.",
+							   pg_rusage_show(&ru0))));
+		}
+	}
+
+
+	MemoryContextDelete(private_context);
+
+	return true;
+}
+
+/*
+ * Reindex concurrently for an arbitrary list of index relations
+ * This is called by ReindexRelationConcurrently and
+ */
+static List *
+ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds, int options,
+		MemoryContext private_context)
+{
+	List	   *newIndexIds = NIL;
+	List	   *relationLocks = NIL;
+	List	   *lockTags = NIL;
+
+	ListCell   *lc,
+			   *lc2;
+
+	MemoryContext oldcontext;
+
+	const int	progress_index[] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_ACCESS_METHOD_OID
+	};
+	int64		progress_vals[4];
+
 	/*-----
 	 * Now we have all the indexes we want to process in indexIds.
 	 *
@@ -3913,42 +4050,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 	/* Start a new transaction to finish process properly */
 	StartTransactionCommand();
 
-	/* Log what we did */
-	if ((params->options & REINDEXOPT_VERBOSE) != 0)
-	{
-		if (relkind == RELKIND_INDEX)
-			ereport(INFO,
-					(errmsg("index \"%s.%s\" was reindexed",
-							relationNamespace, relationName),
-					 errdetail("%s.",
-							   pg_rusage_show(&ru0))));
-		else
-		{
-			foreach(lc, newIndexIds)
-			{
-				ReindexIndexInfo *idx = lfirst(lc);
-				Oid			indOid = idx->indexId;
-
-				ereport(INFO,
-						(errmsg("index \"%s.%s\" was reindexed",
-								get_namespace_name(get_rel_namespace(indOid)),
-								get_rel_name(indOid))));
-				/* Don't show rusage here, since it's not per index. */
-			}
-
-			ereport(INFO,
-					(errmsg("table \"%s.%s\" was reindexed",
-							relationNamespace, relationName),
-					 errdetail("%s.",
-							   pg_rusage_show(&ru0))));
-		}
-	}
-
-	MemoryContextDelete(private_context);
-
 	pgstat_progress_end_command();
 
-	return true;
+	return newIndexIds;
 }
 
 /*
diff --git a/src/test/regress/expected/create_index.out b/src/test/regress/expected/create_index.out
index fc6afab58a..4a03ab2abb 100644
--- a/src/test/regress/expected/create_index.out
+++ b/src/test/regress/expected/create_index.out
@@ -2470,12 +2470,12 @@ COMMIT;
 REINDEX TABLE CONCURRENTLY pg_class; -- no catalog relation
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_class_oid_index; -- no catalog index
-ERROR:  cannot reindex system catalogs concurrently
+ERROR:  concurrent index creation on system catalog tables is not supported
 -- These are the toast table and index of pg_authid.
 REINDEX TABLE CONCURRENTLY pg_toast.pg_toast_1260; -- no catalog toast table
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_toast.pg_toast_1260_index; -- no catalog toast index
-ERROR:  cannot reindex system catalogs concurrently
+ERROR:  concurrent index creation on system catalog tables is not supported
 REINDEX SYSTEM CONCURRENTLY postgres; -- not allowed for SYSTEM
 ERROR:  cannot reindex system catalogs concurrently
 -- Warns about catalog relations
-- 
2.17.0

From fa128228d21fce54babf6f736e833eea3d2f82a3 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sun, 1 Nov 2020 13:46:18 -0600
Subject: [PATCH v12 5/5] More refactoring

---
 src/backend/commands/indexcmds.c           | 183 +++++++++------------
 src/test/regress/expected/create_index.out |   4 +-
 2 files changed, 83 insertions(+), 104 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 08d44a1999..4586942960 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -102,8 +102,8 @@ static void ReindexMultipleInternal(List *relids,
 									ReindexParams *params);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
-static List *ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
-										int options, MemoryContext private_context);
+static List *ReindexIndexesConcurrently(List *indexIds, int options,
+										MemoryContext private_context);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
 
@@ -2625,8 +2625,8 @@ ReindexIndex(RangeVar *indexRelation, ReindexParams *params, bool isTopLevel)
 			.indexId = indOid,
 			/* other fields set later */
 		};
+
 		ReindexIndexesConcurrently(list_make1(&idxinfo),
-				list_make1_oid(IndexGetRelation(indOid, false)),
 				params->options, CurrentMemoryContext);
 	}
 	else
@@ -2961,12 +2961,11 @@ reindex_error_callback(void *arg)
 
 
 /*
- * Given a list of index oids, return a list of leaf partitions by removing
- * any intermediate parents.  heaprels is populated with the corresponding
- * tables.
+ * Given a list of index oids, return a new list of leaf partitions by
+ * excluding any intermediate parents.
  */
 static List *
-leaf_partitions(List *inhoids, int options, List **heaprels)
+leaf_partitions(List *inhoids, int options)
 {
 	List		*partitions = NIL;
 	ListCell	*lc;
@@ -2998,7 +2997,6 @@ leaf_partitions(List *inhoids, int options, List **heaprels)
 
 		/* Save partition OID in current MemoryContext */
 		partitions = lappend_oid(partitions, partoid);
-		*heaprels = lappend_oid(*heaprels, tableoid);
 	}
 
 	return partitions;
@@ -3010,13 +3008,11 @@ leaf_partitions(List *inhoids, int options, List **heaprels)
  *
  * Reindex a set of partitions, per the partitioned index or table given
  * by the caller.
- * XXX: should be further refactored with logic from ReindexRelationConcurrently
  */
 static void
 ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 {
-	List	   *partitions = NIL,
-			*heaprels = NIL;
+	List	   *partitions = NIL;
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
@@ -3070,7 +3066,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	if (relkind == RELKIND_PARTITIONED_INDEX)
 	{
 		old_context = MemoryContextSwitchTo(reindex_context);
-		partitions = leaf_partitions(inhoids, params->options, &heaprels);
+		partitions = leaf_partitions(inhoids, params->options);
 		MemoryContextSwitchTo(old_context);
 	} else {
 		/* Loop over parent tables */
@@ -3083,7 +3079,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 			parttable = table_open(partoid, ShareLock);
 			old_context = MemoryContextSwitchTo(reindex_context);
 			partindexes = RelationGetIndexList(parttable);
-			partindexes = leaf_partitions(partindexes, params->options, &heaprels);
+			partindexes = leaf_partitions(partindexes, params->options);
 			partitions = list_concat(partitions, partindexes);
 
 			MemoryContextSwitchTo(old_context);
@@ -3092,10 +3088,9 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	}
 
 	if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
-		relkind == RELKIND_PARTITIONED_INDEX &&
 		get_rel_persistence(relid) != RELPERSISTENCE_TEMP)
 	{
-		List			   *idxinfos = NIL;
+		List			    *idxinfos = NIL;
 		ReindexIndexInfo	*idxinfo;
 
 		old_context = MemoryContextSwitchTo(reindex_context);
@@ -3110,7 +3105,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		MemoryContextSwitchTo(old_context);
 
 		/* Process all indexes in a single loop */
-		ReindexIndexesConcurrently(idxinfos, heaprels, params->options, reindex_context);
+		ReindexIndexesConcurrently(idxinfos, params->options, reindex_context);
 	} else {
 		/*
 		 * Process each partition listed in a separate transaction.  Note that
@@ -3262,7 +3257,6 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 static bool
 ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 {
-	List	   *heapRelationIds = NIL;
 	List	   *indexIds = NIL;
 	List	   *newIndexIds = NIL;
 	ListCell   *lc,
@@ -3315,14 +3309,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 				 */
 				Relation	heapRelation;
 
-				/* Save the list of relation OIDs in private context */
-				oldcontext = MemoryContextSwitchTo(private_context);
-
-				/* Track this relation for session locks */
-				heapRelationIds = lappend_oid(heapRelationIds, relationOid);
-
-				MemoryContextSwitchTo(oldcontext);
-
 				if (IsCatalogRelationOid(relationOid))
 					ereport(ERROR,
 							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -3335,7 +3321,7 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 												  ShareUpdateExclusiveLock);
 					/* leave if relation does not exist */
 					if (!heapRelation)
-						break;
+						break; // XXX: lremove
 				}
 				else
 					heapRelation = table_open(relationOid,
@@ -3386,14 +3372,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 					Relation	toastRelation = table_open(toastOid,
 														   ShareUpdateExclusiveLock);
 
-					/* Save the list of relation OIDs in private context */
-					oldcontext = MemoryContextSwitchTo(private_context);
-
-					/* Track this relation for session locks */
-					heapRelationIds = lappend_oid(heapRelationIds, toastOid);
-
-					MemoryContextSwitchTo(oldcontext);
-
 					foreach(lc2, RelationGetIndexList(toastRelation))
 					{
 						Oid			cellOid = lfirst_oid(lc2);
@@ -3434,70 +3412,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 				break;
 			}
 		case RELKIND_INDEX:
-			{
-				Oid			heapId = IndexGetRelation(relationOid,
-													  (params->options & REINDEXOPT_MISSING_OK) != 0);
-				Relation	heapRelation;
-				ReindexIndexInfo *idx;
-
-				/* if relation is missing, leave */
-				if (!OidIsValid(heapId))
-					break;
-
-				if (IsCatalogRelationOid(heapId))
-					ereport(ERROR,
-							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-							 errmsg("cannot reindex system catalogs concurrently")));
-
-				/*
-				 * Don't allow reindex for an invalid index on TOAST table, as
-				 * if rebuilt it would not be possible to drop it.  Match
-				 * error message in reindex_index().
-				 */
-				if (IsToastNamespace(get_rel_namespace(relationOid)) &&
-					!get_index_isvalid(relationOid))
-					ereport(ERROR,
-							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-							 errmsg("cannot reindex invalid index on TOAST table")));
-
-				/*
-				 * Check if parent relation can be locked and if it exists,
-				 * this needs to be done at this stage as the list of indexes
-				 * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
-				 * should not be used once all the session locks are taken.
-				 */
-				if ((params->options & REINDEXOPT_MISSING_OK) != 0)
-				{
-					heapRelation = try_table_open(heapId,
-												  ShareUpdateExclusiveLock);
-					/* leave if relation does not exist */
-					if (!heapRelation)
-						break;
-				}
-				else
-					heapRelation = table_open(heapId,
-											  ShareUpdateExclusiveLock);
-				table_close(heapRelation, NoLock);
-
-				/* Save the list of relation OIDs in private context */
-				oldcontext = MemoryContextSwitchTo(private_context);
-
-				/* Track the heap relation of this index for session locks */
-				heapRelationIds = list_make1_oid(heapId);
-
-				/*
-				 * Save the list of relation OIDs in private context.  Note
-				 * that invalid indexes are allowed here.
-				 */
-				idx = palloc(sizeof(ReindexIndexInfo));
-				idx->indexId = relationOid;
-				indexIds = lappend(indexIds, idx);
-				/* other fields set later */
-
-				MemoryContextSwitchTo(oldcontext);
-				break;
-			}
-
 		case RELKIND_PARTITIONED_TABLE:
 		case RELKIND_PARTITIONED_INDEX:
 		default:
@@ -3521,10 +3435,8 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		return false;
 	}
 
-	Assert(heapRelationIds != NIL);
-
 	/* Do the work */
-	newIndexIds = ReindexIndexesConcurrently(indexIds, heapRelationIds, params->options, private_context);
+	newIndexIds = ReindexIndexesConcurrently(indexIds, params->options, private_context);
 
 	/* Log what we did */
 	if ((params->options & REINDEXOPT_VERBOSE) != 0)
@@ -3566,9 +3478,10 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
  * This is called by ReindexRelationConcurrently and
  */
 static List *
-ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds, int options,
+ReindexIndexesConcurrently(List *indexIds, int options,
 		MemoryContext private_context)
 {
+	List		*heapRelationIds = NIL;
 	List	   *newIndexIds = NIL;
 	List	   *relationLocks = NIL;
 	List	   *lockTags = NIL;
@@ -3586,6 +3499,72 @@ ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds, int options,
 	};
 	int64		progress_vals[4];
 
+	foreach(lc, indexIds)
+	{
+		ReindexIndexInfo	*idx = lfirst(lc);
+		Oid			indexrelid = idx->indexId;
+		Oid			heapId = IndexGetRelation(indexrelid,
+											  (options & REINDEXOPT_MISSING_OK) != 0);
+		Relation	heapRelation;
+
+		/* if relation is missing, leave */
+		if (!OidIsValid(heapId))
+			break; // XXX: ldelete?
+
+		if (IsCatalogRelationOid(heapId))
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot reindex system catalogs concurrently")));
+
+		/*
+		 * Don't allow reindex for an invalid index on TOAST table, as
+		 * if rebuilt it would not be possible to drop it.  Match
+		 * error message in reindex_index().
+		 */
+		if (IsToastNamespace(get_rel_namespace(indexrelid)) &&
+			!get_index_isvalid(indexrelid))
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot reindex invalid index on TOAST table")));
+
+		if (IsToastNamespace(get_rel_namespace(indexrelid)) &&
+			!get_index_isvalid(indexrelid))
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot reindex invalid index on TOAST table")));
+
+		/*
+		 * Check if parent relation can be locked and if it exists,
+		 * this needs to be done at this stage as the list of indexes
+		 * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
+		 * should not be used once all the session locks are taken.
+		 */
+		if ((options & REINDEXOPT_MISSING_OK) != 0)
+		{
+			heapRelation = try_table_open(heapId,
+										  ShareUpdateExclusiveLock);
+			/* leave if relation does not exist */
+			if (!heapRelation)
+				break; // ldelete
+		}
+		else
+			heapRelation = table_open(heapId,
+									  ShareUpdateExclusiveLock);
+		table_close(heapRelation, NoLock);
+
+		/* Save the list of relation OIDs in private context */
+		oldcontext = MemoryContextSwitchTo(private_context);
+
+		/* Track the heap relation of this index for session locks */
+		heapRelationIds = lappend_oid(heapRelationIds, heapId);
+		// heapRelationIds = list_make1_oid(heapId);
+
+		/* Note that invalid indexes are allowed here. */
+
+		MemoryContextSwitchTo(oldcontext);
+		// break;
+	}
+
 	/*-----
 	 * Now we have all the indexes we want to process in indexIds.
 	 *
diff --git a/src/test/regress/expected/create_index.out b/src/test/regress/expected/create_index.out
index 4a03ab2abb..fc6afab58a 100644
--- a/src/test/regress/expected/create_index.out
+++ b/src/test/regress/expected/create_index.out
@@ -2470,12 +2470,12 @@ COMMIT;
 REINDEX TABLE CONCURRENTLY pg_class; -- no catalog relation
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_class_oid_index; -- no catalog index
-ERROR:  concurrent index creation on system catalog tables is not supported
+ERROR:  cannot reindex system catalogs concurrently
 -- These are the toast table and index of pg_authid.
 REINDEX TABLE CONCURRENTLY pg_toast.pg_toast_1260; -- no catalog toast table
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_toast.pg_toast_1260_index; -- no catalog toast index
-ERROR:  concurrent index creation on system catalog tables is not supported
+ERROR:  cannot reindex system catalogs concurrently
 REINDEX SYSTEM CONCURRENTLY postgres; -- not allowed for SYSTEM
 ERROR:  cannot reindex system catalogs concurrently
 -- Warns about catalog relations
-- 
2.17.0

From c846ddfc287bfeddb9b389de1869aadf7173c068 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH v13 01/18] Allow CREATE INDEX CONCURRENTLY on partitioned
 table

Note, this effectively reverts 050098b14, so take care to not reintroduce the
bug it fixed.
---
 doc/src/sgml/ref/create_index.sgml     |   9 --
 src/backend/commands/indexcmds.c       | 143 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 ++++++++++-
 src/test/regress/sql/indexing.sql      |  18 +++-
 4 files changed, 176 insertions(+), 54 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index a5271a9f8f..6869a18968 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -686,15 +686,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 127ba7835d..4ac1dacd7d 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -680,17 +681,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1128,6 +1118,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1183,6 +1178,14 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
 			bool		invalidate_parent = false;
@@ -1193,8 +1196,10 @@ DefineIndex(Oid relationId,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1237,10 +1242,12 @@ DefineIndex(Oid relationId,
 					continue;
 				}
 
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				childidxs = RelationGetIndexList(childrel);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1311,10 +1318,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1366,10 +1377,18 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
 
@@ -1379,41 +1398,33 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
-	if (!concurrent)
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
+		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1617,6 +1628,62 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY
+	};
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		/* XXX: need to retrofit progress reporting into it */
+		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 // npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 * XXX: since this is done in 2*N transactions, it could just as well
+	 * call ReindexRelationConcurrently directly
+	 */
+	ReindexMultipleInternal(partitions, &params);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index c93f4470c9..f04abc6897 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c..3d4b6e9bc9 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.17.0

From 6cbf4b5f0e3f795e0253896b13d22a67e513b602 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sun, 14 Feb 2021 18:31:43 -0600
Subject: [PATCH v13 02/18] f! progress reporting

---
 src/backend/commands/indexcmds.c | 33 +++++++-------------------------
 1 file changed, 7 insertions(+), 26 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 4ac1dacd7d..d6567ec231 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1638,40 +1638,20 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 		.options = REINDEXOPT_CONCURRENTLY
 	};
 
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
 	PreventInTransactionBlock(true, "REINDEX INDEX");
 
-	foreach (lc, childs)
+	foreach (lc, find_inheritance_children(indexRelationId, ShareLock))
 	{
 		Oid			partoid = lfirst_oid(lc);
 
-		/* XXX: need to retrofit progress reporting into it */
-		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 // npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
+		if (!get_index_isvalid(partoid) &&
+				RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			ReindexRelationConcurrently(partoid, &params);
 
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
+		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 npart++);
 	}
 
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 * XXX: since this is done in 2*N transactions, it could just as well
-	 * call ReindexRelationConcurrently directly
-	 */
-	ReindexMultipleInternal(partitions, &params);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
@@ -1680,6 +1660,7 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * partitions.
 	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
-- 
2.17.0

From 6dff5512b07735e51c2c7f82cce5b3bd8b9a6faa Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 16:23:02 -0500
Subject: [PATCH v13 03/18] WIP: Add SKIPVALID flag for more integration

XXX: this breaks progress reporting?
---
 src/backend/commands/indexcmds.c | 36 +++++++++++++++-----------------
 src/include/catalog/index.h      |  1 +
 2 files changed, 18 insertions(+), 19 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index d6567ec231..10d4da136f 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1628,40 +1628,33 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
 	};
 
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, find_inheritance_children(indexRelationId, ShareLock))
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		if (!get_index_isvalid(partoid) &&
-				RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			ReindexRelationConcurrently(partoid, &params);
-
-		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 npart++);
-	}
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId, &params, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -3085,6 +3078,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip invalid indexes, if requested */
+		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index e22d506436..994fe94fa1 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.17.0

From 04d6b45f488d67400f7e6d7beb73ad92c00083e3 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 23:52:31 -0500
Subject: [PATCH v13 04/18] ReindexPartitions() to set indisvalid..

Something like this should probably have been included in
a6642b3ae060976b42830b7dc8f29ec190ab05e4

See also 71a05b223, which mentioned the absence of any way to validate an
index.
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 10d4da136f..b21555b6cf 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1643,8 +1643,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1654,9 +1652,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * this commits and then starts a new transaction immediately.
 	 */
 	ReindexPartitions(indexRelationId, &params, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -3098,6 +3093,24 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, params);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.17.0

From 0ba6f83c6df1140037016eb100f8c85dd13070fe Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sun, 1 Nov 2020 12:25:15 -0600
Subject: [PATCH v13 05/18] Refactor to allow reindexing all index partitions
 at once..

The utility of this is to reindex N partitions in 2 transactions, rather than
2*N transactions.
---
 src/backend/commands/indexcmds.c           | 258 ++++++++++++++-------
 src/test/regress/expected/create_index.out |   4 +-
 2 files changed, 182 insertions(+), 80 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index b21555b6cf..f5fea14ff4 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -102,6 +102,8 @@ static void ReindexMultipleInternal(List *relids,
 									ReindexParams *params);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
+static List *ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
+		ReindexParams *params, MemoryContext private_context);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
 
@@ -124,6 +126,15 @@ typedef struct ReindexErrorInfo
 	char		relkind;
 } ReindexErrorInfo;
 
+/* Argument to ReindexIndexConcurrently takes a List* of these */
+typedef struct ReindexIndexInfo
+{
+	Oid			indexId;
+	Oid			tableId;
+	Oid			amId;
+	bool		safe;		/* for set_indexsafe_procflags */
+} ReindexIndexInfo;
+
 /*
  * CheckIndexCompatible
  *		Determine whether an existing index definition is compatible with a
@@ -2636,7 +2647,15 @@ ReindexIndex(RangeVar *indexRelation, ReindexParams *params, bool isTopLevel)
 		ReindexPartitions(indOid, params, isTopLevel);
 	else if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
 			 persistence != RELPERSISTENCE_TEMP)
-		ReindexRelationConcurrently(indOid, params);
+	{
+		ReindexIndexInfo idxinfo = {
+			.indexId = indOid,
+			/* other fields set later */
+		};
+		ReindexIndexesConcurrently(list_make1(&idxinfo),
+				list_make1_oid(IndexGetRelation(indOid, false)),
+				params, CurrentMemoryContext);
+	}
 	else
 	{
 		ReindexParams newparams = *params;
@@ -3002,20 +3021,69 @@ reindex_error_callback(void *arg)
 				   errinfo->relnamespace, errinfo->relname);
 }
 
+
+/*
+ * Given a list of index oids, return a list of leaf partitions by removing
+ * any intermediate parents.  heaprels is populated with the corresponding
+ * tables.
+ */
+static List *
+leaf_indexes(List *inhoids, int options, List **heaprels)
+{
+	List		*partitions = NIL;
+	ListCell	*lc;
+
+	foreach(lc, inhoids)
+	{
+		Oid			partoid = lfirst_oid(lc);
+		Oid			tableoid;
+		Relation	table;
+		char		partkind = get_rel_relkind(partoid);
+
+		/*
+		 * This discards partitioned indexes and foreign tables.
+		 */
+		if (!RELKIND_HAS_STORAGE(partkind))
+			continue;
+
+		Assert(partkind == RELKIND_INDEX);
+
+		/* Skip invalid indexes, if requested */
+		if ((options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
+		/* (try to) Open the table, with lock */
+		tableoid = IndexGetRelation(partoid, false);
+		table = table_open(tableoid, ShareLock);
+		table_close(table, NoLock);
+
+		/* Save partition OID in current MemoryContext */
+		partitions = lappend_oid(partitions, partoid);
+		*heaprels = lappend_oid(*heaprels, tableoid);
+	}
+
+	return partitions;
+}
+
+
 /*
  * ReindexPartitions
  *
  * Reindex a set of partitions, per the partitioned index or table given
  * by the caller.
+ * XXX: should be further refactored with logic from ReindexRelationConcurrently
  */
 static void
 ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 {
-	List	   *partitions = NIL;
+	List	   *partitions = NIL,
+			*heaprels = NIL;
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
 	MemoryContext reindex_context;
+	MemoryContext old_context;
 	List	   *inhoids;
 	ListCell   *lc;
 	ErrorContextCallback errcallback;
@@ -3060,38 +3128,58 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 * The list of relations to reindex are the physical partitions of the
 	 * tree so discard any partitioned table or index.
 	 */
-	foreach(lc, inhoids)
-	{
-		Oid			partoid = lfirst_oid(lc);
-		char		partkind = get_rel_relkind(partoid);
-		MemoryContext old_context;
-
-		/*
-		 * This discards partitioned tables, partitioned indexes and foreign
-		 * tables.
-		 */
-		if (!RELKIND_HAS_STORAGE(partkind))
-			continue;
 
-		/* Skip invalid indexes, if requested */
-		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
-				get_index_isvalid(partoid))
-			continue;
+	if (relkind == RELKIND_PARTITIONED_INDEX)
+	{
+		old_context = MemoryContextSwitchTo(reindex_context);
+		partitions = leaf_indexes(inhoids, params->options, &heaprels);
+		MemoryContextSwitchTo(old_context);
+	} else {
+		/* Loop over parent tables */
+		foreach(lc, inhoids)
+		{
+			Oid		partoid = lfirst_oid(lc);
+			Relation parttable;
+			List	*partindexes;
+
+			parttable = table_open(partoid, ShareLock);
+			old_context = MemoryContextSwitchTo(reindex_context);
+			partindexes = RelationGetIndexList(parttable);
+			partindexes = leaf_indexes(partindexes, params->options, &heaprels);
+			partitions = list_concat(partitions, partindexes);
+
+			MemoryContextSwitchTo(old_context);
+			table_close(parttable, ShareLock);
+		}
+	}
 
-		Assert(partkind == RELKIND_INDEX ||
-			   partkind == RELKIND_RELATION);
+	if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
+		relkind == RELKIND_PARTITIONED_INDEX &&
+		get_rel_persistence(relid) != RELPERSISTENCE_TEMP)
+	{
+		List			   *idxinfos = NIL;
+		ReindexIndexInfo	*idxinfo;
 
-		/* Save partition OID */
 		old_context = MemoryContextSwitchTo(reindex_context);
-		partitions = lappend_oid(partitions, partoid);
+		foreach (lc, partitions)
+		{
+			Oid partoid = lfirst_oid(lc);
+			idxinfo = palloc(sizeof(ReindexIndexInfo));
+			idxinfo->indexId = partoid;
+			/* other fields set later */
+			idxinfos = lappend(idxinfos, idxinfo);
+		}
 		MemoryContextSwitchTo(old_context);
-	}
 
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 */
-	ReindexMultipleInternal(partitions, params);
+		/* Process all indexes in a single loop */
+		ReindexIndexesConcurrently(idxinfos, heaprels, params, reindex_context);
+	} else {
+		/*
+		 * Process each partition listed in a separate transaction.  Note that
+		 * this commits and then starts a new transaction immediately.
+		 */
+		ReindexMultipleInternal(partitions, params);
+	}
 
 	/*
 	 * If indexes exist on all of the partitioned table's children, and we
@@ -3254,18 +3342,9 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 static bool
 ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 {
-	typedef struct ReindexIndexInfo
-	{
-		Oid			indexId;
-		Oid			tableId;
-		Oid			amId;
-		bool		safe;		/* for set_indexsafe_procflags */
-	} ReindexIndexInfo;
 	List	   *heapRelationIds = NIL;
 	List	   *indexIds = NIL;
 	List	   *newIndexIds = NIL;
-	List	   *relationLocks = NIL;
-	List	   *lockTags = NIL;
 	ListCell   *lc,
 			   *lc2;
 	MemoryContext private_context;
@@ -3274,13 +3353,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 	char	   *relationName = NULL;
 	char	   *relationNamespace = NULL;
 	PGRUsage	ru0;
-	const int	progress_index[] = {
-		PROGRESS_CREATEIDX_COMMAND,
-		PROGRESS_CREATEIDX_PHASE,
-		PROGRESS_CREATEIDX_INDEX_OID,
-		PROGRESS_CREATEIDX_ACCESS_METHOD_OID
-	};
-	int64		progress_vals[4];
 
 	/*
 	 * Create a memory context that will survive forced transaction commits we
@@ -3553,6 +3625,69 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 	Assert(heapRelationIds != NIL);
 
+	/* Do the work */
+	newIndexIds = ReindexIndexesConcurrently(indexIds, heapRelationIds, params, private_context);
+
+	/* Log what we did */
+	if ((params->options & REINDEXOPT_VERBOSE) != 0)
+	{
+		if (relkind == RELKIND_INDEX)
+			ereport(INFO,
+					(errmsg("index \"%s.%s\" was reindexed",
+							relationNamespace, relationName),
+					 errdetail("%s.",
+							   pg_rusage_show(&ru0))));
+		else
+		{
+			foreach(lc, newIndexIds)
+			{
+				Oid			indOid = lfirst_oid(lc);
+				ereport(INFO,
+						(errmsg("index \"%s.%s\" was reindexed",
+								get_namespace_name(get_rel_namespace(indOid)),
+								get_rel_name(indOid))));
+				/* Don't show rusage here, since it's not per index. */
+			}
+
+			ereport(INFO,
+					(errmsg("table \"%s.%s\" was reindexed",
+							relationNamespace, relationName),
+					 errdetail("%s.",
+							   pg_rusage_show(&ru0))));
+		}
+	}
+
+
+	MemoryContextDelete(private_context);
+
+	return true;
+}
+
+/*
+ * Reindex concurrently for an arbitrary list of index relations
+ * This is called by ReindexRelationConcurrently and
+ */
+static List *
+ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
+		ReindexParams *params, MemoryContext private_context)
+{
+	List	   *newIndexIds = NIL;
+	List	   *relationLocks = NIL;
+	List	   *lockTags = NIL;
+
+	ListCell   *lc,
+			   *lc2;
+
+	MemoryContext oldcontext;
+
+	const int	progress_index[] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_ACCESS_METHOD_OID
+	};
+	int64		progress_vals[4];
+
 	/*-----
 	 * Now we have all the indexes we want to process in indexIds.
 	 *
@@ -4026,42 +4161,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 	/* Start a new transaction to finish process properly */
 	StartTransactionCommand();
 
-	/* Log what we did */
-	if ((params->options & REINDEXOPT_VERBOSE) != 0)
-	{
-		if (relkind == RELKIND_INDEX)
-			ereport(INFO,
-					(errmsg("index \"%s.%s\" was reindexed",
-							relationNamespace, relationName),
-					 errdetail("%s.",
-							   pg_rusage_show(&ru0))));
-		else
-		{
-			foreach(lc, newIndexIds)
-			{
-				ReindexIndexInfo *idx = lfirst(lc);
-				Oid			indOid = idx->indexId;
-
-				ereport(INFO,
-						(errmsg("index \"%s.%s\" was reindexed",
-								get_namespace_name(get_rel_namespace(indOid)),
-								get_rel_name(indOid))));
-				/* Don't show rusage here, since it's not per index. */
-			}
-
-			ereport(INFO,
-					(errmsg("table \"%s.%s\" was reindexed",
-							relationNamespace, relationName),
-					 errdetail("%s.",
-							   pg_rusage_show(&ru0))));
-		}
-	}
-
-	MemoryContextDelete(private_context);
-
 	pgstat_progress_end_command();
 
-	return true;
+	return newIndexIds;
 }
 
 /*
diff --git a/src/test/regress/expected/create_index.out b/src/test/regress/expected/create_index.out
index 830fdddf24..6f41adf736 100644
--- a/src/test/regress/expected/create_index.out
+++ b/src/test/regress/expected/create_index.out
@@ -2470,12 +2470,12 @@ COMMIT;
 REINDEX TABLE CONCURRENTLY pg_class; -- no catalog relation
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_class_oid_index; -- no catalog index
-ERROR:  cannot reindex system catalogs concurrently
+ERROR:  concurrent index creation on system catalog tables is not supported
 -- These are the toast table and index of pg_authid.
 REINDEX TABLE CONCURRENTLY pg_toast.pg_toast_1260; -- no catalog toast table
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_toast.pg_toast_1260_index; -- no catalog toast index
-ERROR:  cannot reindex system catalogs concurrently
+ERROR:  concurrent index creation on system catalog tables is not supported
 REINDEX SYSTEM CONCURRENTLY postgres; -- not allowed for SYSTEM
 ERROR:  cannot reindex system catalogs concurrently
 -- Warns about catalog relations
-- 
2.17.0

From edde23839fe1dafffe804b24f9068c0f294c0651 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sun, 1 Nov 2020 13:46:18 -0600
Subject: [PATCH v13 06/18] More refactoring

---
 src/backend/commands/indexcmds.c           | 201 ++++++++++-----------
 src/test/regress/expected/create_index.out |   4 +-
 2 files changed, 93 insertions(+), 112 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index f5fea14ff4..5c5596cd28 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -102,8 +102,8 @@ static void ReindexMultipleInternal(List *relids,
 									ReindexParams *params);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
-static List *ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
-		ReindexParams *params, MemoryContext private_context);
+static List *ReindexIndexesConcurrently(List *indexIds, ReindexParams *params,
+		MemoryContext private_context);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
 
@@ -2652,8 +2652,8 @@ ReindexIndex(RangeVar *indexRelation, ReindexParams *params, bool isTopLevel)
 			.indexId = indOid,
 			/* other fields set later */
 		};
+
 		ReindexIndexesConcurrently(list_make1(&idxinfo),
-				list_make1_oid(IndexGetRelation(indOid, false)),
 				params, CurrentMemoryContext);
 	}
 	else
@@ -3023,12 +3023,11 @@ reindex_error_callback(void *arg)
 
 
 /*
- * Given a list of index oids, return a list of leaf partitions by removing
- * any intermediate parents.  heaprels is populated with the corresponding
- * tables.
+ * Given a list of index oids, return a new list of leaf partitions by
+ * excluding any intermediate parents.
  */
 static List *
-leaf_indexes(List *inhoids, int options, List **heaprels)
+leaf_indexes(List *inhoids, int options)
 {
 	List		*partitions = NIL;
 	ListCell	*lc;
@@ -3060,7 +3059,6 @@ leaf_indexes(List *inhoids, int options, List **heaprels)
 
 		/* Save partition OID in current MemoryContext */
 		partitions = lappend_oid(partitions, partoid);
-		*heaprels = lappend_oid(*heaprels, tableoid);
 	}
 
 	return partitions;
@@ -3072,13 +3070,11 @@ leaf_indexes(List *inhoids, int options, List **heaprels)
  *
  * Reindex a set of partitions, per the partitioned index or table given
  * by the caller.
- * XXX: should be further refactored with logic from ReindexRelationConcurrently
  */
 static void
 ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 {
-	List	   *partitions = NIL,
-			*heaprels = NIL;
+	List	   *partitions = NIL;
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
@@ -3132,7 +3128,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	if (relkind == RELKIND_PARTITIONED_INDEX)
 	{
 		old_context = MemoryContextSwitchTo(reindex_context);
-		partitions = leaf_indexes(inhoids, params->options, &heaprels);
+		partitions = leaf_indexes(inhoids, params->options);
 		MemoryContextSwitchTo(old_context);
 	} else {
 		/* Loop over parent tables */
@@ -3145,7 +3141,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 			parttable = table_open(partoid, ShareLock);
 			old_context = MemoryContextSwitchTo(reindex_context);
 			partindexes = RelationGetIndexList(parttable);
-			partindexes = leaf_indexes(partindexes, params->options, &heaprels);
+			partindexes = leaf_indexes(partindexes, params->options);
 			partitions = list_concat(partitions, partindexes);
 
 			MemoryContextSwitchTo(old_context);
@@ -3154,10 +3150,9 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	}
 
 	if ((params->options & REINDEXOPT_CONCURRENTLY) != 0 &&
-		relkind == RELKIND_PARTITIONED_INDEX &&
 		get_rel_persistence(relid) != RELPERSISTENCE_TEMP)
 	{
-		List			   *idxinfos = NIL;
+		List			    *idxinfos = NIL;
 		ReindexIndexInfo	*idxinfo;
 
 		old_context = MemoryContextSwitchTo(reindex_context);
@@ -3172,7 +3167,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		MemoryContextSwitchTo(old_context);
 
 		/* Process all indexes in a single loop */
-		ReindexIndexesConcurrently(idxinfos, heaprels, params, reindex_context);
+		ReindexIndexesConcurrently(idxinfos, params, reindex_context);
 	} else {
 		/*
 		 * Process each partition listed in a separate transaction.  Note that
@@ -3342,7 +3337,6 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 static bool
 ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 {
-	List	   *heapRelationIds = NIL;
 	List	   *indexIds = NIL;
 	List	   *newIndexIds = NIL;
 	ListCell   *lc,
@@ -3395,14 +3389,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 				 */
 				Relation	heapRelation;
 
-				/* Save the list of relation OIDs in private context */
-				oldcontext = MemoryContextSwitchTo(private_context);
-
-				/* Track this relation for session locks */
-				heapRelationIds = lappend_oid(heapRelationIds, relationOid);
-
-				MemoryContextSwitchTo(oldcontext);
-
 				if (IsCatalogRelationOid(relationOid))
 					ereport(ERROR,
 							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -3415,7 +3401,7 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 												  ShareUpdateExclusiveLock);
 					/* leave if relation does not exist */
 					if (!heapRelation)
-						break;
+						break; // XXX: lremove
 				}
 				else
 					heapRelation = table_open(relationOid,
@@ -3473,14 +3459,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 					Relation	toastRelation = table_open(toastOid,
 														   ShareUpdateExclusiveLock);
 
-					/* Save the list of relation OIDs in private context */
-					oldcontext = MemoryContextSwitchTo(private_context);
-
-					/* Track this relation for session locks */
-					heapRelationIds = lappend_oid(heapRelationIds, toastOid);
-
-					MemoryContextSwitchTo(oldcontext);
-
 					foreach(lc2, RelationGetIndexList(toastRelation))
 					{
 						Oid			cellOid = lfirst_oid(lc2);
@@ -3521,78 +3499,6 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 				break;
 			}
 		case RELKIND_INDEX:
-			{
-				Oid			heapId = IndexGetRelation(relationOid,
-													  (params->options & REINDEXOPT_MISSING_OK) != 0);
-				Relation	heapRelation;
-				ReindexIndexInfo *idx;
-
-				/* if relation is missing, leave */
-				if (!OidIsValid(heapId))
-					break;
-
-				if (IsCatalogRelationOid(heapId))
-					ereport(ERROR,
-							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-							 errmsg("cannot reindex system catalogs concurrently")));
-
-				/*
-				 * Don't allow reindex for an invalid index on TOAST table, as
-				 * if rebuilt it would not be possible to drop it.  Match
-				 * error message in reindex_index().
-				 */
-				if (IsToastNamespace(get_rel_namespace(relationOid)) &&
-					!get_index_isvalid(relationOid))
-					ereport(ERROR,
-							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-							 errmsg("cannot reindex invalid index on TOAST table")));
-
-				/*
-				 * Check if parent relation can be locked and if it exists,
-				 * this needs to be done at this stage as the list of indexes
-				 * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
-				 * should not be used once all the session locks are taken.
-				 */
-				if ((params->options & REINDEXOPT_MISSING_OK) != 0)
-				{
-					heapRelation = try_table_open(heapId,
-												  ShareUpdateExclusiveLock);
-					/* leave if relation does not exist */
-					if (!heapRelation)
-						break;
-				}
-				else
-					heapRelation = table_open(heapId,
-											  ShareUpdateExclusiveLock);
-
-				if (OidIsValid(params->tablespaceOid) &&
-					IsSystemRelation(heapRelation))
-					ereport(ERROR,
-							(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-							 errmsg("cannot move system relation \"%s\"",
-									get_rel_name(relationOid))));
-
-				table_close(heapRelation, NoLock);
-
-				/* Save the list of relation OIDs in private context */
-				oldcontext = MemoryContextSwitchTo(private_context);
-
-				/* Track the heap relation of this index for session locks */
-				heapRelationIds = list_make1_oid(heapId);
-
-				/*
-				 * Save the list of relation OIDs in private context.  Note
-				 * that invalid indexes are allowed here.
-				 */
-				idx = palloc(sizeof(ReindexIndexInfo));
-				idx->indexId = relationOid;
-				indexIds = lappend(indexIds, idx);
-				/* other fields set later */
-
-				MemoryContextSwitchTo(oldcontext);
-				break;
-			}
-
 		case RELKIND_PARTITIONED_TABLE:
 		case RELKIND_PARTITIONED_INDEX:
 		default:
@@ -3623,10 +3529,10 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 				 errmsg("cannot move non-shared relation to tablespace \"%s\"",
 						get_tablespace_name(params->tablespaceOid))));
 
-	Assert(heapRelationIds != NIL);
+	// Assert(heapRelationIds != NIL);
 
 	/* Do the work */
-	newIndexIds = ReindexIndexesConcurrently(indexIds, heapRelationIds, params, private_context);
+	newIndexIds = ReindexIndexesConcurrently(indexIds, params, private_context);
 
 	/* Log what we did */
 	if ((params->options & REINDEXOPT_VERBOSE) != 0)
@@ -3668,9 +3574,10 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
  * This is called by ReindexRelationConcurrently and
  */
 static List *
-ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
-		ReindexParams *params, MemoryContext private_context)
+ReindexIndexesConcurrently(List *indexIds, ReindexParams *params,
+		MemoryContext private_context)
 {
+	List		*heapRelationIds = NIL;
 	List	   *newIndexIds = NIL;
 	List	   *relationLocks = NIL;
 	List	   *lockTags = NIL;
@@ -3688,6 +3595,80 @@ ReindexIndexesConcurrently(List *indexIds, List *heapRelationIds,
 	};
 	int64		progress_vals[4];
 
+	/* It's not a shared catalog, so refuse to move it to shared tablespace */
+	if (params->tablespaceOid == GLOBALTABLESPACE_OID && false)
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("cannot move non-shared relation to tablespace \"%s\"",
+						get_tablespace_name(params->tablespaceOid))));
+
+	foreach(lc, indexIds)
+	{
+		ReindexIndexInfo	*idx = lfirst(lc);
+		Oid			indexrelid = idx->indexId;
+		Oid			heapId = IndexGetRelation(indexrelid,
+											  (params->options & REINDEXOPT_MISSING_OK) != 0);
+		Relation	heapRelation;
+
+		/* if relation is missing, leave */
+		if (!OidIsValid(heapId))
+			break; // XXX: ldelete?
+
+		if (IsCatalogRelationOid(heapId))
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot reindex system catalogs concurrently")));
+
+		/*
+		 * Don't allow reindex for an invalid index on TOAST table, as
+		 * if rebuilt it would not be possible to drop it.  Match
+		 * error message in reindex_index().
+		 */
+		if (IsToastNamespace(get_rel_namespace(indexrelid)) &&
+			!get_index_isvalid(indexrelid))
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot reindex invalid index on TOAST table")));
+
+		if (OidIsValid(params->tablespaceOid) &&
+			IsCatalogRelationOid(indexrelid))
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("cannot move system relation \"%s\"",
+							get_rel_name(indexrelid))));
+
+		/*
+		 * Check if parent relation can be locked and if it exists,
+		 * this needs to be done at this stage as the list of indexes
+		 * to rebuild is not complete yet, and REINDEXOPT_MISSING_OK
+		 * should not be used once all the session locks are taken.
+		 */
+		if ((params->options & REINDEXOPT_MISSING_OK) != 0)
+		{
+			heapRelation = try_table_open(heapId,
+										  ShareUpdateExclusiveLock);
+			/* leave if relation does not exist */
+			if (!heapRelation)
+				break; // ldelete
+		}
+		else
+			heapRelation = table_open(heapId,
+									  ShareUpdateExclusiveLock);
+		table_close(heapRelation, NoLock);
+
+		/* Save the list of relation OIDs in private context */
+		oldcontext = MemoryContextSwitchTo(private_context);
+
+		/* Track the heap relation of this index for session locks */
+		heapRelationIds = lappend_oid(heapRelationIds, heapId);
+		// heapRelationIds = list_make1_oid(heapId);
+
+		/* Note that invalid indexes are allowed here. */
+
+		MemoryContextSwitchTo(oldcontext);
+		// break;
+	}
+
 	/*-----
 	 * Now we have all the indexes we want to process in indexIds.
 	 *
diff --git a/src/test/regress/expected/create_index.out b/src/test/regress/expected/create_index.out
index 6f41adf736..830fdddf24 100644
--- a/src/test/regress/expected/create_index.out
+++ b/src/test/regress/expected/create_index.out
@@ -2470,12 +2470,12 @@ COMMIT;
 REINDEX TABLE CONCURRENTLY pg_class; -- no catalog relation
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_class_oid_index; -- no catalog index
-ERROR:  concurrent index creation on system catalog tables is not supported
+ERROR:  cannot reindex system catalogs concurrently
 -- These are the toast table and index of pg_authid.
 REINDEX TABLE CONCURRENTLY pg_toast.pg_toast_1260; -- no catalog toast table
 ERROR:  cannot reindex system catalogs concurrently
 REINDEX INDEX CONCURRENTLY pg_toast.pg_toast_1260_index; -- no catalog toast index
-ERROR:  concurrent index creation on system catalog tables is not supported
+ERROR:  cannot reindex system catalogs concurrently
 REINDEX SYSTEM CONCURRENTLY postgres; -- not allowed for SYSTEM
 ERROR:  cannot reindex system catalogs concurrently
 -- Warns about catalog relations
-- 
2.17.0

From fb60da3c0fac8f1699a6caeea57476770c66576d Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH 1/5] Allow CREATE INDEX CONCURRENTLY on partitioned table

Note, this effectively reverts 050098b14, so take care to not reintroduce the
bug it fixed.
---
 doc/src/sgml/ref/create_index.sgml     |   9 --
 src/backend/commands/indexcmds.c       | 143 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 ++++++++++-
 src/test/regress/sql/indexing.sql      |  18 +++-
 4 files changed, 176 insertions(+), 54 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 965dcf472c..7c75119d78 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -686,15 +686,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 8bc652ecd3..9ab1a66971 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -680,17 +681,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1128,6 +1118,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1183,6 +1178,14 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
 			bool		invalidate_parent = false;
@@ -1193,8 +1196,10 @@ DefineIndex(Oid relationId,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1237,10 +1242,12 @@ DefineIndex(Oid relationId,
 					continue;
 				}
 
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				childidxs = RelationGetIndexList(childrel);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1311,10 +1318,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1366,10 +1377,18 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
 
@@ -1379,41 +1398,33 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
-	if (!concurrent)
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
+		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1626,6 +1637,62 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY
+	};
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		/* XXX: need to retrofit progress reporting into it */
+		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 // npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 * XXX: since this is done in 2*N transactions, it could just as well
+	 * call ReindexRelationConcurrently directly
+	 */
+	ReindexMultipleInternal(partitions, &params);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index c93f4470c9..f04abc6897 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c..3d4b6e9bc9 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.17.0

From 0dfaaf20b8333b816ab2d4501675e9a7dd0fc436 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sun, 14 Feb 2021 18:31:43 -0600
Subject: [PATCH 2/5] f! progress reporting

---
 src/backend/commands/indexcmds.c | 33 +++++++-------------------------
 1 file changed, 7 insertions(+), 26 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 9ab1a66971..8f4eab22eb 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1647,40 +1647,20 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 		.options = REINDEXOPT_CONCURRENTLY
 	};
 
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
 	PreventInTransactionBlock(true, "REINDEX INDEX");
 
-	foreach (lc, childs)
+	foreach (lc, find_inheritance_children(indexRelationId, ShareLock))
 	{
 		Oid			partoid = lfirst_oid(lc);
 
-		/* XXX: need to retrofit progress reporting into it */
-		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 // npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
+		if (!get_index_isvalid(partoid) &&
+				RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			ReindexRelationConcurrently(partoid, &params);
 
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
+		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 npart++);
 	}
 
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 * XXX: since this is done in 2*N transactions, it could just as well
-	 * call ReindexRelationConcurrently directly
-	 */
-	ReindexMultipleInternal(partitions, &params);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
@@ -1689,6 +1669,7 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * partitions.
 	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
-- 
2.17.0

From 2a24e8bbc6f23d3e688d72a652ba0bd0dd2dc4af Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 16:23:02 -0500
Subject: [PATCH 3/5] WIP: Add SKIPVALID flag for more integration

XXX: this breaks progress reporting?
---
 src/backend/commands/indexcmds.c | 36 +++++++++++++++-----------------
 src/include/catalog/index.h      |  1 +
 2 files changed, 18 insertions(+), 19 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 8f4eab22eb..e54314e9a4 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1637,40 +1637,33 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
 	};
 
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, find_inheritance_children(indexRelationId, ShareLock))
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		if (!get_index_isvalid(partoid) &&
-				RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			ReindexRelationConcurrently(partoid, &params);
-
-		pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 npart++);
-	}
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId, &params, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -3094,6 +3087,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip valid indexes, if requested */
+		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index e22d506436..994fe94fa1 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.17.0

From 2b6fedc7cc9f3a0dfe522d789ffed414a799605c Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Fri, 30 Oct 2020 23:52:31 -0500
Subject: [PATCH 4/5] ReindexPartitions() to set indisvalid..

Something like this should probably have been included in
a6642b3ae060976b42830b7dc8f29ec190ab05e4

See also 71a05b223, which mentioned the absence of any way to validate an
index.
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index e54314e9a4..99508b0d36 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1652,8 +1652,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1663,9 +1661,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * this commits and then starts a new transaction immediately.
 	 */
 	ReindexPartitions(indexRelationId, &params, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -3107,6 +3102,24 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, params);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.17.0

From eaad7c3ed2fda93fdb91aea60294f60489444bf7 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:28:42 +0300
Subject: [PATCH 1/4] Allow CREATE INDEX CONCURRENTLY on partitioned table

0001-Allow-CREATE-INDEX-CONCURRENTLY-on-partitioned-table.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com

Fixes:
  - rel was used after table_close();
  - it seems childidxs shouldn't live in ind_context;
  - updated doc.
---
 doc/src/sgml/ref/create_index.sgml     |  14 +--
 src/backend/commands/indexcmds.c       | 151 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 +++++++++-
 src/test/regress/sql/indexing.sql      |  18 ++-
 4 files changed, 186 insertions(+), 57 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 91eaaabc90f..ffa98692430 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -641,7 +641,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while creating index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -688,15 +691,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 560dcc87a2c..666ced8e1d7 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -670,17 +671,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1119,6 +1109,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1174,18 +1169,30 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel, true);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
 			bool		invalidate_parent = false;
 			TupleDesc	parentDesc;
 			Oid		   *opfamOids;
+			char		*relname;
 
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			relname = pstrdup(RelationGetRelationName(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1220,18 +1227,21 @@ DefineIndex(Oid relationId,
 						ereport(ERROR,
 								(errcode(ERRCODE_WRONG_OBJECT_TYPE),
 								 errmsg("cannot create unique index on partitioned table \"%s\"",
-										RelationGetRelationName(rel)),
+										relname),
 								 errdetail("Table \"%s\" contains partitions that are foreign tables.",
-										   RelationGetRelationName(rel))));
+										   relname)));
 
 					table_close(childrel, lockmode);
 					continue;
 				}
 
 				childidxs = RelationGetIndexList(childrel);
+
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1302,10 +1312,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1357,12 +1371,21 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
+			pfree(relname);
 
 			/*
 			 * The pg_index row we inserted for this index was marked
@@ -1370,41 +1393,33 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
-	if (!concurrent)
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
+		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1617,6 +1632,62 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY
+	};
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		/* XXX: need to retrofit progress reporting into it */
+		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 // npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 * XXX: since this is done in 2*N transactions, it could just as well
+	 * call ReindexRelationConcurrently directly
+	 */
+	ReindexMultipleInternal(partitions, &params);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 193f7801912..a4ccae50de3 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c2..3d4b6e9bc95 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.25.1

From 91bed69a737ca73fb5f79725e7bffc31e617b61b Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:31:40 +0300
Subject: [PATCH 2/4] Add SKIPVALID flag for more integration

Combined
0002-f-progress-reporting.patch and
0003-WIP-Add-SKIPVALID-flag-for-more-integration.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com
---
 src/backend/commands/indexcmds.c | 57 ++++++++++----------------------
 src/include/catalog/index.h      |  1 +
 2 files changed, 19 insertions(+), 39 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 666ced8e1d7..56e4c0b7575 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1632,59 +1632,33 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
 	};
 
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, childs)
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		/* XXX: need to retrofit progress reporting into it */
-		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 // npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
-
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
-	}
-
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 * XXX: since this is done in 2*N transactions, it could just as well
-	 * call ReindexRelationConcurrently directly
-	 */
-	ReindexMultipleInternal(partitions, &params);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId, &params, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -3106,6 +3080,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip valid indexes, if requested */
+		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index a1d6e3b645f..c31b66ad0b9 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.25.1

From 1b9cb0f8f32c1863a52c5a3fd53103537a938ebd Mon Sep 17 00:00:00 2001
From: Alexander Pyhalov <a.pyhalov@postgrespro.ru>
Date: Tue, 8 Feb 2022 21:15:05 +0300
Subject: [PATCH 3/4] Try to fix create index progress report

---
 src/backend/commands/indexcmds.c | 67 ++++++++++++++++++++++++++------
 src/include/catalog/index.h      |  1 +
 2 files changed, 57 insertions(+), 11 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 56e4c0b7575..b237c30bc80 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -99,11 +99,14 @@ static void reindex_error_callback(void *args);
 static void ReindexPartitions(Oid relid, ReindexParams *params,
 							  bool isTopLevel);
 static void ReindexMultipleInternal(List *relids,
-									ReindexParams *params);
+									ReindexParams *params,
+									Oid parent,
+									int npart);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
+static void report_create_partition_index_done(Oid parent, int npart);
 
 /*
  * callback argument type for RangeVarCallbackForReindexIndex()
@@ -1184,6 +1187,7 @@ DefineIndex(Oid relationId,
 			Oid		   *opfamOids;
 			char		*relname;
 
+
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
@@ -1640,7 +1644,7 @@ static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID | REINDEXOPT_REPORT_CREATE_PART
 	};
 
 	/*
@@ -1653,6 +1657,8 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
 
+	pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE, PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN);
+
 	/*
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
@@ -2987,7 +2993,7 @@ ReindexMultipleTables(const char *objectName, ReindexObjectType objectKind,
 	 * Process each relation listed in a separate transaction.  Note that this
 	 * commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(relids, params);
+	ReindexMultipleInternal(relids, params, InvalidOid, 0);
 
 	MemoryContextDelete(private_context);
 }
@@ -3023,6 +3029,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
+	int			npart = 1;
 	MemoryContext reindex_context;
 	List	   *inhoids;
 	ListCell   *lc;
@@ -3083,7 +3090,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		/* Skip valid indexes, if requested */
 		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
 				get_index_isvalid(partoid))
+		{
+			if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+				report_create_partition_index_done(relid, npart++);
 			continue;
+		}
 
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
@@ -3098,7 +3109,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(partitions, params);
+	ReindexMultipleInternal(partitions, params, relid, npart);
 
 	/*
 	 * Clean up working storage --- note we must do this after
@@ -3116,7 +3127,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
  * and starts a new transaction when finished.
  */
 static void
-ReindexMultipleInternal(List *relids, ReindexParams *params)
+ReindexMultipleInternal(List *relids, ReindexParams *params, Oid parent, int npart)
 {
 	ListCell   *l;
 
@@ -3210,6 +3221,9 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 		}
 
 		CommitTransactionCommand();
+
+		if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+			report_create_partition_index_done(parent, npart++);
 	}
 
 	StartTransactionCommand();
@@ -3592,7 +3606,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
 			elog(ERROR, "cannot reindex a temporary table concurrently");
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		/* Don't overwrite CREATE INDEX command */
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  idx->tableId);
 
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
@@ -3745,9 +3761,11 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved.  Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
 		progress_vals[2] = newidx->indexId;
@@ -3809,10 +3827,12 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved. Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
 		progress_vals[2] = newidx->indexId;
@@ -4047,7 +4067,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 	MemoryContextDelete(private_context);
 
-	pgstat_progress_end_command();
+	/* Don't overwrite CREATE INDEX command. */
+	if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		pgstat_progress_end_command();
 
 	return true;
 }
@@ -4183,6 +4205,29 @@ IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
 	}
 }
 
+/*
+ * Update pgstat progress report to indicate that create index on
+ * partition was finished.
+ */
+static void
+report_create_partition_index_done(Oid index, int npart)
+{
+	const int   progress_cols[] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_PARTITIONS_DONE
+	};
+	const int64 progress_vals[] = {
+		PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY,
+		index,
+		PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
+		npart
+	};
+
+	pgstat_progress_update_multi_param(4, progress_cols, progress_vals);
+}
+
 /*
  * Subroutine of IndexSetParentIndex to update the relispartition flag of the
  * given index to the given value.
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index c31b66ad0b9..b5b0a71e7d4 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -43,6 +43,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
 #define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
+#define REINDEXOPT_REPORT_CREATE_PART	0x20	/* report that index was created for partition */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.25.1

From 62f2816188d3d38fe0d76d3212d6f551a59c0ae7 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:39:59 +0300
Subject: [PATCH 4/4] ReindexPartitions() to set indisvalid

0004-ReindexPartitions-to-set-indisvalid.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index b237c30bc80..355a7626549 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1651,8 +1651,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1664,9 +1662,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * this commits and then starts a new transaction immediately.
 	 */
 	ReindexPartitions(indexRelationId, &params, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -3111,6 +3106,24 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, params, relid, npart);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.25.1

From 18fa3c27a3311294a7abfdc0674ef6143c65423b Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:28:42 +0300
Subject: [PATCH 1/5] Allow CREATE INDEX CONCURRENTLY on partitioned table

0001-Allow-CREATE-INDEX-CONCURRENTLY-on-partitioned-table.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com

Fixes:
  - rel was used after table_close();
  - it seems childidxs shouldn't live in ind_context;
  - updated doc.
---
 doc/src/sgml/ref/create_index.sgml     |  14 +--
 src/backend/commands/indexcmds.c       | 151 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 +++++++++-
 src/test/regress/sql/indexing.sql      |  18 ++-
 4 files changed, 186 insertions(+), 57 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 91eaaabc90f..ffa98692430 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -641,7 +641,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while creating index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -688,15 +691,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index cd30f15eba6..a34a1b133a0 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -670,17 +671,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1119,6 +1109,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1174,18 +1169,30 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel, true);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
 			bool		invalidate_parent = false;
 			TupleDesc	parentDesc;
 			Oid		   *opfamOids;
+			char		*relname;
 
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			relname = pstrdup(RelationGetRelationName(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1220,18 +1227,21 @@ DefineIndex(Oid relationId,
 						ereport(ERROR,
 								(errcode(ERRCODE_WRONG_OBJECT_TYPE),
 								 errmsg("cannot create unique index on partitioned table \"%s\"",
-										RelationGetRelationName(rel)),
+										relname),
 								 errdetail("Table \"%s\" contains partitions that are foreign tables.",
-										   RelationGetRelationName(rel))));
+										   relname)));
 
 					table_close(childrel, lockmode);
 					continue;
 				}
 
 				childidxs = RelationGetIndexList(childrel);
+
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1302,10 +1312,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1357,12 +1371,21 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
+			pfree(relname);
 
 			/*
 			 * The pg_index row we inserted for this index was marked
@@ -1370,41 +1393,33 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
-	if (!concurrent)
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
+		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1617,6 +1632,62 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY
+	};
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		/* XXX: need to retrofit progress reporting into it */
+		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 // npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 * XXX: since this is done in 2*N transactions, it could just as well
+	 * call ReindexRelationConcurrently directly
+	 */
+	ReindexMultipleInternal(partitions, &params);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 193f7801912..a4ccae50de3 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c2..3d4b6e9bc95 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.25.1

From 1cba217c9293a4dac1b08451000442733c8c78cd Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:31:40 +0300
Subject: [PATCH 2/5] Add SKIPVALID flag for more integration

Combined
0002-f-progress-reporting.patch and
0003-WIP-Add-SKIPVALID-flag-for-more-integration.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com
---
 src/backend/commands/indexcmds.c | 57 ++++++++++----------------------
 src/include/catalog/index.h      |  1 +
 2 files changed, 19 insertions(+), 39 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index a34a1b133a0..090e792ff47 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1632,59 +1632,33 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
 	};
 
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, childs)
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		/* XXX: need to retrofit progress reporting into it */
-		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 // npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
-
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
-	}
-
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 * XXX: since this is done in 2*N transactions, it could just as well
-	 * call ReindexRelationConcurrently directly
-	 */
-	ReindexMultipleInternal(partitions, &params);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId, &params, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -3106,6 +3080,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip valid indexes, if requested */
+		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index a1d6e3b645f..c31b66ad0b9 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.25.1

From 1a4e873d247dffc1b1af16a546f0d28714d3fb9b Mon Sep 17 00:00:00 2001
From: Alexander Pyhalov <a.pyhalov@postgrespro.ru>
Date: Tue, 8 Feb 2022 21:15:05 +0300
Subject: [PATCH 3/5] Try to fix create index progress report

---
 src/backend/commands/indexcmds.c | 67 ++++++++++++++++++++++++++------
 src/include/catalog/index.h      |  1 +
 2 files changed, 57 insertions(+), 11 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 090e792ff47..57df92985fe 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -99,11 +99,14 @@ static void reindex_error_callback(void *args);
 static void ReindexPartitions(Oid relid, ReindexParams *params,
 							  bool isTopLevel);
 static void ReindexMultipleInternal(List *relids,
-									ReindexParams *params);
+									ReindexParams *params,
+									Oid parent,
+									int npart);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
+static void report_create_partition_index_done(Oid parent, int npart);
 
 /*
  * callback argument type for RangeVarCallbackForReindexIndex()
@@ -1184,6 +1187,7 @@ DefineIndex(Oid relationId,
 			Oid		   *opfamOids;
 			char		*relname;
 
+
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
@@ -1640,7 +1644,7 @@ static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID | REINDEXOPT_REPORT_CREATE_PART
 	};
 
 	/*
@@ -1653,6 +1657,8 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
 
+	pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE, PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN);
+
 	/*
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
@@ -2987,7 +2993,7 @@ ReindexMultipleTables(const char *objectName, ReindexObjectType objectKind,
 	 * Process each relation listed in a separate transaction.  Note that this
 	 * commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(relids, params);
+	ReindexMultipleInternal(relids, params, InvalidOid, 0);
 
 	MemoryContextDelete(private_context);
 }
@@ -3023,6 +3029,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
+	int			npart = 1;
 	MemoryContext reindex_context;
 	List	   *inhoids;
 	ListCell   *lc;
@@ -3083,7 +3090,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		/* Skip valid indexes, if requested */
 		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
 				get_index_isvalid(partoid))
+		{
+			if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+				report_create_partition_index_done(relid, npart++);
 			continue;
+		}
 
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
@@ -3098,7 +3109,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(partitions, params);
+	ReindexMultipleInternal(partitions, params, relid, npart);
 
 	/*
 	 * Clean up working storage --- note we must do this after
@@ -3116,7 +3127,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
  * and starts a new transaction when finished.
  */
 static void
-ReindexMultipleInternal(List *relids, ReindexParams *params)
+ReindexMultipleInternal(List *relids, ReindexParams *params, Oid parent, int npart)
 {
 	ListCell   *l;
 
@@ -3210,6 +3221,9 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 		}
 
 		CommitTransactionCommand();
+
+		if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+			report_create_partition_index_done(parent, npart++);
 	}
 
 	StartTransactionCommand();
@@ -3592,7 +3606,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
 			elog(ERROR, "cannot reindex a temporary table concurrently");
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		/* Don't overwrite CREATE INDEX command */
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  idx->tableId);
 
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
@@ -3745,9 +3761,11 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved.  Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
 		progress_vals[2] = newidx->indexId;
@@ -3809,10 +3827,12 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved. Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
 		progress_vals[2] = newidx->indexId;
@@ -4047,7 +4067,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 	MemoryContextDelete(private_context);
 
-	pgstat_progress_end_command();
+	/* Don't overwrite CREATE INDEX command. */
+	if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		pgstat_progress_end_command();
 
 	return true;
 }
@@ -4183,6 +4205,29 @@ IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
 	}
 }
 
+/*
+ * Update pgstat progress report to indicate that create index on
+ * partition was finished.
+ */
+static void
+report_create_partition_index_done(Oid index, int npart)
+{
+	const int   progress_cols[] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_PARTITIONS_DONE
+	};
+	const int64 progress_vals[] = {
+		PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY,
+		index,
+		PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
+		npart
+	};
+
+	pgstat_progress_update_multi_param(4, progress_cols, progress_vals);
+}
+
 /*
  * Subroutine of IndexSetParentIndex to update the relispartition flag of the
  * given index to the given value.
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index c31b66ad0b9..b5b0a71e7d4 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -43,6 +43,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
 #define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
+#define REINDEXOPT_REPORT_CREATE_PART	0x20	/* report that index was created for partition */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.25.1

From 44055a8f644d153a6df919eba02db41f8085b9e8 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:39:59 +0300
Subject: [PATCH 4/5] ReindexPartitions() to set indisvalid

0004-ReindexPartitions-to-set-indisvalid.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 57df92985fe..21f1ceaea63 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1651,8 +1651,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1664,9 +1662,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * this commits and then starts a new transaction immediately.
 	 */
 	ReindexPartitions(indexRelationId, &params, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -3111,6 +3106,24 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, params, relid, npart);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.25.1

From 75db7db9fae509c64609315b86aa85706e710e2b Mon Sep 17 00:00:00 2001
From: Alexander Pyhalov <a.pyhalov@postgrespro.ru>
Date: Mon, 28 Feb 2022 10:50:58 +0300
Subject: [PATCH 5/5] Mark intermediate partitioned indexes as valid

---
 src/backend/commands/indexcmds.c       | 33 ++++++++++-
 src/test/regress/expected/indexing.out | 80 +++++++++++++++++++++++++-
 src/test/regress/sql/indexing.sql      |  8 +++
 3 files changed, 118 insertions(+), 3 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 21f1ceaea63..fabb5b14898 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -3021,6 +3021,7 @@ static void
 ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 {
 	List	   *partitions = NIL;
+	List	   *inhpartindexes = NIL;
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
@@ -3075,6 +3076,17 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		char		partkind = get_rel_relkind(partoid);
 		MemoryContext old_context;
 
+		/* Create a list of invalid inherited partitioned indexes */
+		if (partkind == RELKIND_PARTITIONED_INDEX)
+		{
+			if (partoid == relid || get_index_isvalid(partoid))
+				continue;
+
+			old_context = MemoryContextSwitchTo(reindex_context);
+			inhpartindexes = lappend_oid(inhpartindexes, partoid);
+			MemoryContextSwitchTo(old_context);
+		}
+
 		/*
 		 * This discards partitioned tables, partitioned indexes and foreign
 		 * tables.
@@ -3119,9 +3131,28 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		Oid	tableoid = IndexGetRelation(relid, false);
 		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
 
-		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		/*
+		 * Both lists include their parent relation as well as any
+		 * intermediate partitioned rels
+		 */
 		if (list_length(inhoids) == list_length(child_tables))
+		{
 			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+
+			/* Mark any intermediate partitioned index as valid */
+			foreach(lc, inhpartindexes)
+			{
+				Oid         partoid = lfirst_oid(lc);
+
+				Assert(get_rel_relkind(partoid) == RELKIND_PARTITIONED_INDEX);
+				Assert(!get_index_isvalid(partoid));
+
+				/* Can't mark an index valid without marking it ready */
+				index_set_state_flags(partoid, INDEX_CREATE_SET_READY);
+				CommandCounterIncrement();
+				index_set_state_flags(partoid, INDEX_CREATE_SET_VALID);
+			}
+		}
 	}
 
 	/*
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index a4ccae50de3..b4f1aea6fca 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -57,6 +57,8 @@ create table idxpart11 partition of idxpart1 for values from (0) to (10) partiti
 create table idxpart111 partition of idxpart11 default partition by range(a);
 create table idxpart1111 partition of idxpart111 default partition by range(a);
 create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
 insert into idxpart2 values(10),(10); -- not unique
 create index concurrently on idxpart (a); -- partitioned
 create index concurrently on idxpart1 (a); -- partitioned and partition
@@ -76,7 +78,7 @@ Partition key: RANGE (a)
 Indexes:
     "idxpart_a_idx" btree (a)
     "idxpart_a_idx1" UNIQUE, btree (a) INVALID
-Number of partitions: 2 (Use \d+ to list them.)
+Number of partitions: 3 (Use \d+ to list them.)
 
 \d idxpart1
         Partitioned table "public.idxpart1"
@@ -88,11 +90,59 @@ Number of partitions: 2 (Use \d+ to list them.)
 Partition of: idxpart FOR VALUES FROM (0) TO (10)
 Partition key: RANGE (a)
 Indexes:
-    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx" btree (a)
     "idxpart1_a_idx1" btree (a)
     "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
 Number of partitions: 1 (Use \d+ to list them.)
 
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" btree (a)
+    "idxpart11_a_idx2" btree (a)
+    "idxpart11_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" btree (a)
+    "idxpart111_a_idx2" btree (a)
+    "idxpart111_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" btree (a)
+    "idxpart1111_a_idx2" btree (a)
+    "idxpart1111_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
 \d idxpart2
               Table "public.idxpart2"
  Column |  Type   | Collation | Nullable | Default 
@@ -107,6 +157,32 @@ Indexes:
     "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
     "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
 
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 3d4b6e9bc95..06673c15199 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -36,6 +36,9 @@ create table idxpart11 partition of idxpart1 for values from (0) to (10) partiti
 create table idxpart111 partition of idxpart11 default partition by range(a);
 create table idxpart1111 partition of idxpart111 default partition by range(a);
 create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
 insert into idxpart2 values(10),(10); -- not unique
 create index concurrently on idxpart (a); -- partitioned
 create index concurrently on idxpart1 (a); -- partitioned and partition
@@ -44,7 +47,12 @@ create index concurrently on idxpart2 (a); -- leaf
 create unique index concurrently on idxpart (a); -- partitioned, unique failure
 \d idxpart
 \d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
 \d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.25.1

From 5c11849ceb2a1feb0e44dbdf30cc27de0282a659 Mon Sep 17 00:00:00 2001
From: Alexander Pyhalov <a.pyhalov@postgrespro.ru>
Date: Mon, 28 Feb 2022 10:50:58 +0300
Subject: [PATCH 5/5] Mark intermediate partitioned indexes as valid

---
 src/backend/commands/indexcmds.c       | 33 ++++++++++-
 src/test/regress/expected/indexing.out | 80 +++++++++++++++++++++++++-
 src/test/regress/sql/indexing.sql      |  8 +++
 3 files changed, 118 insertions(+), 3 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index d09f0390413..d3ced6265b6 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -3139,6 +3139,7 @@ static void
 ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 {
 	List	   *partitions = NIL;
+	List	   *inhpartindexes = NIL;
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
@@ -3193,6 +3194,17 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		char		partkind = get_rel_relkind(partoid);
 		MemoryContext old_context;
 
+		/* Create a list of invalid inherited partitioned indexes */
+		if (partkind == RELKIND_PARTITIONED_INDEX)
+		{
+			if (partoid == relid || get_index_isvalid(partoid))
+				continue;
+
+			old_context = MemoryContextSwitchTo(reindex_context);
+			inhpartindexes = lappend_oid(inhpartindexes, partoid);
+			MemoryContextSwitchTo(old_context);
+		}
+
 		/*
 		 * This discards partitioned tables, partitioned indexes and foreign
 		 * tables.
@@ -3237,9 +3249,28 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		Oid	tableoid = IndexGetRelation(relid, false);
 		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
 
-		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		/*
+		 * Both lists include their parent relation as well as any
+		 * intermediate partitioned rels
+		 */
 		if (list_length(inhoids) == list_length(child_tables))
+		{
 			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+
+			/* Mark any intermediate partitioned index as valid */
+			foreach(lc, inhpartindexes)
+			{
+				Oid         partoid = lfirst_oid(lc);
+
+				Assert(get_rel_relkind(partoid) == RELKIND_PARTITIONED_INDEX);
+				Assert(!get_index_isvalid(partoid));
+
+				/* Can't mark an index valid without marking it ready */
+				index_set_state_flags(partoid, INDEX_CREATE_SET_READY);
+				CommandCounterIncrement();
+				index_set_state_flags(partoid, INDEX_CREATE_SET_VALID);
+			}
+		}
 	}
 
 	/*
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index a4ccae50de3..b4f1aea6fca 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -57,6 +57,8 @@ create table idxpart11 partition of idxpart1 for values from (0) to (10) partiti
 create table idxpart111 partition of idxpart11 default partition by range(a);
 create table idxpart1111 partition of idxpart111 default partition by range(a);
 create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
 insert into idxpart2 values(10),(10); -- not unique
 create index concurrently on idxpart (a); -- partitioned
 create index concurrently on idxpart1 (a); -- partitioned and partition
@@ -76,7 +78,7 @@ Partition key: RANGE (a)
 Indexes:
     "idxpart_a_idx" btree (a)
     "idxpart_a_idx1" UNIQUE, btree (a) INVALID
-Number of partitions: 2 (Use \d+ to list them.)
+Number of partitions: 3 (Use \d+ to list them.)
 
 \d idxpart1
         Partitioned table "public.idxpart1"
@@ -88,11 +90,59 @@ Number of partitions: 2 (Use \d+ to list them.)
 Partition of: idxpart FOR VALUES FROM (0) TO (10)
 Partition key: RANGE (a)
 Indexes:
-    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx" btree (a)
     "idxpart1_a_idx1" btree (a)
     "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
 Number of partitions: 1 (Use \d+ to list them.)
 
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" btree (a)
+    "idxpart11_a_idx2" btree (a)
+    "idxpart11_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" btree (a)
+    "idxpart111_a_idx2" btree (a)
+    "idxpart111_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" btree (a)
+    "idxpart1111_a_idx2" btree (a)
+    "idxpart1111_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
 \d idxpart2
               Table "public.idxpart2"
  Column |  Type   | Collation | Nullable | Default 
@@ -107,6 +157,32 @@ Indexes:
     "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
     "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
 
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 3d4b6e9bc95..06673c15199 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -36,6 +36,9 @@ create table idxpart11 partition of idxpart1 for values from (0) to (10) partiti
 create table idxpart111 partition of idxpart11 default partition by range(a);
 create table idxpart1111 partition of idxpart111 default partition by range(a);
 create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
 insert into idxpart2 values(10),(10); -- not unique
 create index concurrently on idxpart (a); -- partitioned
 create index concurrently on idxpart1 (a); -- partitioned and partition
@@ -44,7 +47,12 @@ create index concurrently on idxpart2 (a); -- leaf
 create unique index concurrently on idxpart (a); -- partitioned, unique failure
 \d idxpart
 \d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
 \d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 71838e0146e5150013c48818710d899e69786dc0 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:39:59 +0300
Subject: [PATCH 4/5] ReindexPartitions() to set indisvalid

0004-ReindexPartitions-to-set-indisvalid.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com
---
 src/backend/commands/indexcmds.c | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index b231a05b8c9..d09f0390413 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1716,8 +1716,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
-	 * This must be done only while holding a lock which precludes adding
-	 * partitions.
 	 */
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
@@ -1729,9 +1727,6 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	 * this commits and then starts a new transaction immediately.
 	 */
 	ReindexPartitions(indexRelationId, &params, true);
-
-	CommandCounterIncrement();
-	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
 
 /*
@@ -3229,6 +3224,24 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 */
 	ReindexMultipleInternal(partitions, params, relid, npart);
 
+	/*
+	 * If indexes exist on all of the partitioned table's children, and we
+	 * just reindexed them, then we know they're valid, and so can mark the
+	 * parent index as valid.
+	 * This handles the case of CREATE INDEX CONCURRENTLY.
+	 * See also: validatePartitionedIndex().
+	 */
+	if (get_rel_relkind(relid) == RELKIND_PARTITIONED_INDEX
+			&& !get_index_isvalid(relid))
+	{
+		Oid	tableoid = IndexGetRelation(relid, false);
+		List	*child_tables = find_all_inheritors(tableoid, ShareLock, NULL);
+
+		/* Both lists include their parent relation as well as any intermediate partitioned rels */
+		if (list_length(inhoids) == list_length(child_tables))
+			index_set_state_flags(relid, INDEX_CREATE_SET_VALID);
+	}
+
 	/*
 	 * Clean up working storage --- note we must do this after
 	 * StartTransactionCommand, else we might be trying to delete the active
-- 
2.34.1

From f5b8afe1aafb78c53a527d379b978ce9f1fe06d6 Mon Sep 17 00:00:00 2001
From: Alexander Pyhalov <a.pyhalov@postgrespro.ru>
Date: Tue, 8 Feb 2022 21:15:05 +0300
Subject: [PATCH 3/5] Try to fix create index progress report

---
 src/backend/commands/indexcmds.c | 67 ++++++++++++++++++++++++++------
 src/include/catalog/index.h      |  1 +
 2 files changed, 57 insertions(+), 11 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 149235780b9..b231a05b8c9 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -102,11 +102,14 @@ static void reindex_error_callback(void *args);
 static void ReindexPartitions(Oid relid, ReindexParams *params,
 							  bool isTopLevel);
 static void ReindexMultipleInternal(List *relids,
-									ReindexParams *params);
+									ReindexParams *params,
+									Oid parent,
+									int npart);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
+static void report_create_partition_index_done(Oid parent, int npart);
 
 /*
  * callback argument type for RangeVarCallbackForReindexIndex()
@@ -1220,6 +1223,7 @@ DefineIndex(Oid relationId,
 			Oid		   *opfamOids;
 			char		*relname;
 
+
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
@@ -1705,7 +1709,7 @@ static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID | REINDEXOPT_REPORT_CREATE_PART
 	};
 
 	/*
@@ -1718,6 +1722,8 @@ reindex_invalid_child_indexes(Oid indexRelationId)
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
 
+	pgstat_progress_update_param(PROGRESS_CREATEIDX_PHASE, PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN);
+
 	/*
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
@@ -3105,7 +3111,7 @@ ReindexMultipleTables(const char *objectName, ReindexObjectType objectKind,
 	 * Process each relation listed in a separate transaction.  Note that this
 	 * commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(relids, params);
+	ReindexMultipleInternal(relids, params, InvalidOid, 0);
 
 	MemoryContextDelete(private_context);
 }
@@ -3141,6 +3147,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
+	int			npart = 1;
 	MemoryContext reindex_context;
 	List	   *inhoids;
 	ListCell   *lc;
@@ -3201,7 +3208,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		/* Skip valid indexes, if requested */
 		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
 				get_index_isvalid(partoid))
+		{
+			if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+				report_create_partition_index_done(relid, npart++);
 			continue;
+		}
 
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
@@ -3216,7 +3227,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(partitions, params);
+	ReindexMultipleInternal(partitions, params, relid, npart);
 
 	/*
 	 * Clean up working storage --- note we must do this after
@@ -3234,7 +3245,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
  * and starts a new transaction when finished.
  */
 static void
-ReindexMultipleInternal(List *relids, ReindexParams *params)
+ReindexMultipleInternal(List *relids, ReindexParams *params, Oid parent, int npart)
 {
 	ListCell   *l;
 
@@ -3328,6 +3339,9 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 		}
 
 		CommitTransactionCommand();
+
+		if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+			report_create_partition_index_done(parent, npart++);
 	}
 
 	StartTransactionCommand();
@@ -3723,7 +3737,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
 			elog(ERROR, "cannot reindex a temporary table concurrently");
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		/* Don't overwrite CREATE INDEX command */
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  idx->tableId);
 
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
@@ -3883,9 +3899,11 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved.  Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
 		progress_vals[2] = newidx->indexId;
@@ -3947,10 +3965,12 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved. Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
 		progress_vals[2] = newidx->indexId;
@@ -4185,7 +4205,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 	MemoryContextDelete(private_context);
 
-	pgstat_progress_end_command();
+	/* Don't overwrite CREATE INDEX command. */
+	if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		pgstat_progress_end_command();
 
 	return true;
 }
@@ -4321,6 +4343,29 @@ IndexSetParentIndex(Relation partitionIdx, Oid parentOid)
 	}
 }
 
+/*
+ * Update pgstat progress report to indicate that create index on
+ * partition was finished.
+ */
+static void
+report_create_partition_index_done(Oid index, int npart)
+{
+	const int   progress_cols[] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_PARTITIONS_DONE
+	};
+	const int64 progress_vals[] = {
+		PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY,
+		index,
+		PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
+		npart
+	};
+
+	pgstat_progress_update_multi_param(4, progress_cols, progress_vals);
+}
+
 /*
  * Subroutine of IndexSetParentIndex to update the relispartition flag of the
  * given index to the given value.
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index c31b66ad0b9..b5b0a71e7d4 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -43,6 +43,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
 #define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
+#define REINDEXOPT_REPORT_CREATE_PART	0x20	/* report that index was created for partition */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.34.1

From 4f16e729a835dc8123a8ce593e6ae47542b82c83 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:31:40 +0300
Subject: [PATCH 2/5] Add SKIPVALID flag for more integration

Combined
0002-f-progress-reporting.patch and
0003-WIP-Add-SKIPVALID-flag-for-more-integration.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com
---
 src/backend/commands/indexcmds.c | 57 ++++++++++----------------------
 src/include/catalog/index.h      |  1 +
 2 files changed, 19 insertions(+), 39 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 784f6d9eb87..149235780b9 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -1697,59 +1697,33 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
-/* Reindex invalid child indexes created earlier */
+/*
+ * Reindex invalid child indexes created earlier thereby validating
+ * the parent index.
+ */
 static void
 reindex_invalid_child_indexes(Oid indexRelationId)
 {
-	ListCell *lc;
-	int		npart = 0;
 	ReindexParams params = {
-		.options = REINDEXOPT_CONCURRENTLY
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_SKIPVALID
 	};
 
-	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
-			ALLOCSET_DEFAULT_SIZES);
-	MemoryContext	oldcontext;
-	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
-	List		*partitions = NIL;
-
-	PreventInTransactionBlock(true, "REINDEX INDEX");
-
-	foreach (lc, childs)
-	{
-		Oid			partoid = lfirst_oid(lc);
-
-		/* XXX: need to retrofit progress reporting into it */
-		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-									 // npart++);
-
-		if (get_index_isvalid(partoid) ||
-				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
-			continue;
-
-		/* Save partition OID */
-		oldcontext = MemoryContextSwitchTo(ind_context);
-		partitions = lappend_oid(partitions, partoid);
-		MemoryContextSwitchTo(oldcontext);
-	}
-
-	/*
-	 * Process each partition listed in a separate transaction.  Note that
-	 * this commits and then starts a new transaction immediately.
-	 * XXX: since this is done in 2*N transactions, it could just as well
-	 * call ReindexRelationConcurrently directly
-	 */
-	ReindexMultipleInternal(partitions, &params);
-
 	/*
 	 * CIC needs to mark a partitioned index as VALID, which itself
 	 * requires setting READY, which is unset for CIC (even though
 	 * it's meaningless for an index without storage).
 	 * This must be done only while holding a lock which precludes adding
 	 * partitions.
-	 * See also: validatePartitionedIndex().
 	 */
+	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 */
+	ReindexPartitions(indexRelationId, &params, true);
+
 	CommandCounterIncrement();
 	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
 }
@@ -3224,6 +3198,11 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 		if (!RELKIND_HAS_STORAGE(partkind))
 			continue;
 
+		/* Skip valid indexes, if requested */
+		if ((params->options & REINDEXOPT_SKIPVALID) != 0 &&
+				get_index_isvalid(partoid))
+			continue;
+
 		Assert(partkind == RELKIND_INDEX ||
 			   partkind == RELKIND_RELATION);
 
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index a1d6e3b645f..c31b66ad0b9 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_SKIPVALID	0x10	/* skip valid indexes */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.34.1

From f4933625541b030fabcb15d1426f496598949898 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 7 Feb 2022 10:28:42 +0300
Subject: [PATCH 1/5] Allow CREATE INDEX CONCURRENTLY on partitioned table

0001-Allow-CREATE-INDEX-CONCURRENTLY-on-partitioned-table.patch from
https://www.postgresql.org/message-id/20210226182019.GU20769@telsasoft.com

Fixes:
  - rel was used after table_close();
  - it seems childidxs shouldn't live in ind_context;
  - updated doc.
---
 doc/src/sgml/ref/create_index.sgml     |  14 +--
 src/backend/commands/indexcmds.c       | 151 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out |  60 +++++++++-
 src/test/regress/sql/indexing.sql      |  18 ++-
 4 files changed, 186 insertions(+), 57 deletions(-)

diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 9ffcdc629e6..2040b27b685 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -642,7 +642,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while creating index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -689,15 +692,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 99f5ab83c32..784f6d9eb87 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -68,6 +68,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -695,17 +696,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1145,6 +1135,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1210,18 +1205,30 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel, true);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc(sizeof(Oid) * nparts);
 			bool		invalidate_parent = false;
 			TupleDesc	parentDesc;
 			Oid		   *opfamOids;
+			char		*relname;
 
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			relname = pstrdup(RelationGetRelationName(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
-			parentDesc = RelationGetDescr(rel);
 			opfamOids = palloc(sizeof(Oid) * numberOfKeyAttributes);
 			for (i = 0; i < numberOfKeyAttributes; i++)
 				opfamOids[i] = get_opclass_family(classObjectId[i]);
@@ -1265,9 +1272,9 @@ DefineIndex(Oid relationId,
 						ereport(ERROR,
 								(errcode(ERRCODE_WRONG_OBJECT_TYPE),
 								 errmsg("cannot create unique index on partitioned table \"%s\"",
-										RelationGetRelationName(rel)),
+										relname),
 								 errdetail("Table \"%s\" contains partitions that are foreign tables.",
-										   RelationGetRelationName(rel))));
+										   relname)));
 
 					AtEOXact_GUC(false, child_save_nestlevel);
 					SetUserIdAndSecContext(child_save_userid,
@@ -1277,9 +1284,12 @@ DefineIndex(Oid relationId,
 				}
 
 				childidxs = RelationGetIndexList(childrel);
+
+				oldcontext = MemoryContextSwitchTo(ind_context);
 				attmap =
 					build_attrmap_by_name(RelationGetDescr(childrel),
 										  parentDesc);
+				MemoryContextSwitchTo(oldcontext);
 
 				foreach(cell, childidxs)
 				{
@@ -1353,10 +1363,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1417,12 +1431,21 @@ DefineIndex(Oid relationId,
 								skip_build, quiet);
 					SetUserIdAndSecContext(child_save_userid,
 										   child_save_sec_context);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
+			pfree(relname);
 
 			/*
 			 * The pg_index row we inserted for this index was marked
@@ -1430,24 +1453,9 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
@@ -1455,21 +1463,28 @@ DefineIndex(Oid relationId,
 		 */
 		AtEOXact_GUC(false, root_save_nestlevel);
 		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
+		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1682,6 +1697,62 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	int		npart = 0;
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY
+	};
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_inheritance_children(indexRelationId, ShareLock);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+
+		/* XXX: need to retrofit progress reporting into it */
+		// pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+									 // npart++);
+
+		if (get_index_isvalid(partoid) ||
+				!RELKIND_HAS_STORAGE(get_rel_relkind(partoid)))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 * XXX: since this is done in 2*N transactions, it could just as well
+	 * call ReindexRelationConcurrently directly
+	 */
+	ReindexMultipleInternal(partitions, &params);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though
+	 * it's meaningless for an index without storage).
+	 * This must be done only while holding a lock which precludes adding
+	 * partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_READY);
+	CommandCounterIncrement();
+	index_set_state_flags(indexRelationId, INDEX_CREATE_SET_VALID);
+}
 
 /*
  * CheckMutability
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 193f7801912..a4ccae50de3 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,63 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 2 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a) INVALID
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 42f398b67c2..3d4b6e9bc95 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,22 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart2
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 4ba360eaaac5e1ac169d41c26cf6213b0c6a2432 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH] Allow CREATE INDEX CONCURRENTLY on partitioned table

Note, this effectively reverts 050098b14, so take care to not reintroduce the
bug it fixed.
---
 doc/src/sgml/ddl.sgml                  |   4 +-
 doc/src/sgml/ref/create_index.sgml     |   9 --
 src/backend/commands/indexcmds.c       | 214 +++++++++++++++++++------
 src/include/catalog/index.h            |   1 +
 src/test/regress/expected/indexing.out | 136 +++++++++++++++-
 src/test/regress/sql/indexing.sql      |  26 ++-
 6 files changed, 320 insertions(+), 70 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 03c01937094..fd56e21ef49 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4131,9 +4131,7 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      so that they are applied automatically to the entire hierarchy.
      This is very
      convenient, as not only will the existing partitions become indexed, but
-     also any partitions that are created in the future will.  One limitation is
-     that it's not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier when creating such a partitioned index.  To avoid long lock
+     also any partitions that are created in the future will.  To avoid long lock
      times, it is possible to use <command>CREATE INDEX ON ONLY</command>
      the partitioned table; such an index is marked invalid, and the partitions
      do not get the index applied automatically.  The indexes on partitions can
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 40986aa502f..fc8cda655f0 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -692,15 +692,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 91cee27743d..bb98e745267 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -71,6 +71,7 @@
 
 
 /* non-export function prototypes */
+static void reindex_invalid_child_indexes(Oid indexRelationId);
 static bool CompareOpclassOptions(Datum *opts1, Datum *opts2, int natts);
 static void CheckPredicate(Expr *predicate);
 static void ComputeIndexAttrs(IndexInfo *indexInfo,
@@ -104,7 +105,9 @@ static void reindex_error_callback(void *arg);
 static void ReindexPartitions(Oid relid, ReindexParams *params,
 							  bool isTopLevel);
 static void ReindexMultipleInternal(List *relids,
-									ReindexParams *params);
+									ReindexParams *params,
+									Oid parent,
+									int npart);
 static bool ReindexRelationConcurrently(Oid relationOid,
 										ReindexParams *params);
 static void update_relispartition(Oid relationId, bool newval);
@@ -697,17 +700,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1147,6 +1139,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1212,17 +1209,30 @@ DefineIndex(Oid relationId,
 		partdesc = RelationGetPartitionDesc(rel, true);
 		if ((!stmt->relation || stmt->relation->inh) && partdesc->nparts > 0)
 		{
+			/*
+			 * Need to close the relation before recursing into children, so
+			 * copy needed data into a longlived context.
+			 */
+
+			MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+					ALLOCSET_DEFAULT_SIZES);
+			MemoryContext	oldcontext = MemoryContextSwitchTo(ind_context);
 			int			nparts = partdesc->nparts;
 			Oid		   *part_oids = palloc_array(Oid, nparts);
 			bool		invalidate_parent = false;
 			Relation	parentIndex;
 			TupleDesc	parentDesc;
+			char		*relname;
 
 			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_TOTAL,
 										 nparts);
 
 			/* Make a local copy of partdesc->oids[], just for safety */
 			memcpy(part_oids, partdesc->oids, sizeof(Oid) * nparts);
+			parentDesc = CreateTupleDescCopy(RelationGetDescr(rel));
+			relname = pstrdup(RelationGetRelationName(rel));
+			table_close(rel, NoLock);
+			MemoryContextSwitchTo(oldcontext);
 
 			/*
 			 * We'll need an IndexInfo describing the parent index.  The one
@@ -1235,8 +1245,6 @@ DefineIndex(Oid relationId,
 			parentIndex = index_open(indexRelationId, lockmode);
 			indexInfo = BuildIndexInfo(parentIndex);
 
-			parentDesc = RelationGetDescr(rel);
-
 			/*
 			 * For each partition, scan all existing indexes; if one matches
 			 * our index definition and is not already attached to some other
@@ -1276,9 +1284,9 @@ DefineIndex(Oid relationId,
 						ereport(ERROR,
 								(errcode(ERRCODE_WRONG_OBJECT_TYPE),
 								 errmsg("cannot create unique index on partitioned table \"%s\"",
-										RelationGetRelationName(rel)),
+										relname),
 								 errdetail("Table \"%s\" contains partitions that are foreign tables.",
-										   RelationGetRelationName(rel))));
+										   relname)));
 
 					AtEOXact_GUC(false, child_save_nestlevel);
 					SetUserIdAndSecContext(child_save_userid,
@@ -1364,10 +1372,14 @@ DefineIndex(Oid relationId,
 				 */
 				if (!found)
 				{
-					IndexStmt  *childStmt = copyObject(stmt);
+					IndexStmt  *childStmt;
 					bool		found_whole_row;
 					ListCell   *lc;
 
+					oldcontext = MemoryContextSwitchTo(ind_context);
+					childStmt = copyObject(stmt);
+					MemoryContextSwitchTo(oldcontext);
+
 					/*
 					 * We can't use the same index name for the child index,
 					 * so clear idxname to let the recursive invocation choose
@@ -1426,14 +1438,24 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+
 					SetUserIdAndSecContext(child_save_userid,
 										   child_save_sec_context);
+					if (concurrent)
+					{
+						PopActiveSnapshot();
+						PushActiveSnapshot(GetTransactionSnapshot());
+						invalidate_parent = true;
+					}
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
+			pfree(relname);
 
 			index_close(parentIndex, lockmode);
 
@@ -1443,46 +1465,40 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
-		}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
+		} else
+			table_close(rel, NoLock);
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
 		 * done here.
 		 */
+
 		AtEOXact_GUC(false, root_save_nestlevel);
 		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
+
 		if (!OidIsValid(parentIndexId))
+		{
+			if (concurrent)
+				reindex_invalid_child_indexes(indexRelationId);
+
 			pgstat_progress_end_command();
+		}
+
 		return address;
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
-		table_close(rel, NoLock);
+		/*
+		 * We're done if this is the top-level index,
+		 * or the catalog-only phase of a partition built concurrently
+		 */
 
-		/* If this is the top-level index, we're done. */
+		table_close(rel, NoLock);
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1695,6 +1711,70 @@ DefineIndex(Oid relationId,
 	return address;
 }
 
+/* Reindex invalid child indexes created earlier */
+static void
+reindex_invalid_child_indexes(Oid indexRelationId)
+{
+	ListCell *lc;
+	List	*parentindexes = NIL;
+	int	npart = 1;
+	ReindexParams params = {
+		.options = REINDEXOPT_CONCURRENTLY | REINDEXOPT_REPORT_CREATE_PART
+	};
+
+	MemoryContext	ind_context = AllocSetContextCreate(PortalContext, "CREATE INDEX",
+			ALLOCSET_DEFAULT_SIZES);
+	MemoryContext	oldcontext;
+	List		*childs = find_all_inheritors(indexRelationId, ShareLock, NULL);
+	List		*partitions = NIL;
+
+	PreventInTransactionBlock(true, "REINDEX INDEX");
+
+	foreach (lc, childs)
+	{
+		Oid			partoid = lfirst_oid(lc);
+		char		partkind = get_rel_relkind(partoid);
+
+		if (partkind == RELKIND_PARTITIONED_INDEX)
+		{
+			oldcontext = MemoryContextSwitchTo(ind_context);
+			parentindexes = lappend_oid(parentindexes, partoid);
+			MemoryContextSwitchTo(oldcontext);
+		}
+
+		// npart++ ?
+		if (!RELKIND_HAS_STORAGE(partkind) || get_index_isvalid(partoid))
+			continue;
+
+		/* Save partition OID */
+		oldcontext = MemoryContextSwitchTo(ind_context);
+		partitions = lappend_oid(partitions, partoid);
+		MemoryContextSwitchTo(oldcontext);
+	}
+
+	/*
+	 * Process each partition listed in a separate transaction.  Note that
+	 * this commits and then starts a new transaction immediately.
+	 * XXX: since this is done in 2*N transactions, it could just as well
+	 * call ReindexRelationConcurrently directly
+	 */
+	ReindexMultipleInternal(partitions, &params, indexRelationId, npart);
+
+	/*
+	 * CIC needs to mark a partitioned index as VALID, which itself
+	 * requires setting READY, which is unset for CIC (even though it's
+	 * meaningless for an index without storage).  This must be done only
+	 * while holding a lock which precludes adding partitions.
+	 * See also: validatePartitionedIndex().
+	 */
+	foreach (lc, parentindexes)
+	{
+		Oid	partoid = lfirst_oid(lc);
+		index_set_state_flags(partoid, INDEX_CREATE_SET_READY);
+		CommandCounterIncrement();
+		index_set_state_flags(partoid, INDEX_CREATE_SET_VALID);
+	}
+}
 
 /*
  * CheckMutability
@@ -3084,7 +3164,7 @@ ReindexMultipleTables(const char *objectName, ReindexObjectType objectKind,
 	 * Process each relation listed in a separate transaction.  Note that this
 	 * commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(relids, params);
+	ReindexMultipleInternal(relids, params, InvalidOid, 0);
 
 	MemoryContextDelete(private_context);
 }
@@ -3120,6 +3200,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	char		relkind = get_rel_relkind(relid);
 	char	   *relname = get_rel_name(relid);
 	char	   *relnamespace = get_namespace_name(get_rel_namespace(relid));
+	int			npart = 1;
 	MemoryContext reindex_context;
 	List	   *inhoids;
 	ListCell   *lc;
@@ -3190,7 +3271,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(partitions, params);
+	ReindexMultipleInternal(partitions, params, relid, npart);
 
 	/*
 	 * Clean up working storage --- note we must do this after
@@ -3208,7 +3289,7 @@ ReindexPartitions(Oid relid, ReindexParams *params, bool isTopLevel)
  * and starts a new transaction when finished.
  */
 static void
-ReindexMultipleInternal(List *relids, ReindexParams *params)
+ReindexMultipleInternal(List *relids, ReindexParams *params, Oid parent, int npart)
 {
 	ListCell   *l;
 
@@ -3302,6 +3383,29 @@ ReindexMultipleInternal(List *relids, ReindexParams *params)
 		}
 
 		CommitTransactionCommand();
+
+		if (params->options & REINDEXOPT_REPORT_CREATE_PART)
+		{
+			/*
+			 * Update pgstat progress report to indicate that create index on
+			 * partition was finished.
+			 */
+			const int   progress_cols[] = {
+				PROGRESS_CREATEIDX_COMMAND,
+				PROGRESS_CREATEIDX_INDEX_OID,
+				PROGRESS_CREATEIDX_PHASE,
+				PROGRESS_CREATEIDX_PARTITIONS_DONE
+			};
+			const int64 progress_vals[] = {
+				PROGRESS_CREATEIDX_COMMAND_CREATE_CONCURRENTLY,
+				parent,
+				PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN,
+				npart++
+			};
+
+			pgstat_progress_update_multi_param(4, progress_cols, progress_vals);
+		}
+
 	}
 
 	StartTransactionCommand();
@@ -3697,7 +3801,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
 			elog(ERROR, "cannot reindex a temporary table concurrently");
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		/* Don't overwrite CREATE INDEX command */
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  idx->tableId);
 
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
@@ -3857,9 +3963,11 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved.  Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
 		progress_vals[2] = newidx->indexId;
@@ -3921,10 +4029,12 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/*
 		 * Update progress for the index to build, with the correct parent
-		 * table involved.
+		 * table involved. Don't overwrite CREATE INDEX command.
 		 */
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  newidx->tableId);
+
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
 		progress_vals[2] = newidx->indexId;
@@ -4159,7 +4269,9 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 	MemoryContextDelete(private_context);
 
-	pgstat_progress_end_command();
+	/* Don't overwrite CREATE INDEX command. */
+	if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		pgstat_progress_end_command();
 
 	return true;
 }
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index 91c28868d47..4142894c230 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_REPORT_CREATE_PART	0x10	/* report that index was created for partition */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 1bdd430f063..6b2320244b8 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,139 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx2_ccnew"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" btree (a)
+    "idxpart1_a_idx2" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" btree (a)
+    "idxpart11_a_idx2" btree (a)
+    "idxpart11_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" btree (a)
+    "idxpart111_a_idx2" btree (a)
+    "idxpart111_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" btree (a)
+    "idxpart1111_a_idx2" btree (a)
+    "idxpart1111_a_idx3" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" btree (a)
+    "idxpart2_a_idx2" UNIQUE, btree (a) INVALID
+    "idxpart2_a_idx2_ccnew" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 429120e7104..14e0513386c 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart (a); -- partitioned
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart2 (a); -- leaf
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.25.1

From 8eb9fd7ce7d34c5c323c47b60a7f883f360ef090 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Sat, 3 Dec 2022 18:20:03 +0400
Subject: [PATCH] turn off reindex reporting for create

---
 src/backend/commands/indexcmds.c | 36 +++++++++++++++++++-------------
 1 file changed, 21 insertions(+), 15 deletions(-)

diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index a2775931e2..b3c713037f 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -3804,14 +3804,16 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 
 		/* Don't overwrite CREATE INDEX command */
 		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		{
 			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  idx->tableId);
 
-		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
-		progress_vals[1] = 0;	/* initializing */
-		progress_vals[2] = idx->indexId;
-		progress_vals[3] = idx->amId;
-		pgstat_progress_update_multi_param(4, progress_index, progress_vals);
+			progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
+			progress_vals[1] = 0;	/* initializing */
+			progress_vals[2] = idx->indexId;
+			progress_vals[3] = idx->amId;
+			pgstat_progress_update_multi_param(4, progress_index, progress_vals);
+		}
 
 		/* Choose a temporary relation name for the new index */
 		concurrentName = ChooseRelationName(get_rel_name(idx->indexId),
@@ -3967,13 +3969,15 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		 * table involved.  Don't overwrite CREATE INDEX command.
 		 */
 		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		{
 			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, newidx->tableId);
 
-		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
-		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
-		progress_vals[2] = newidx->indexId;
-		progress_vals[3] = newidx->amId;
-		pgstat_progress_update_multi_param(4, progress_index, progress_vals);
+			progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
+			progress_vals[1] = PROGRESS_CREATEIDX_PHASE_BUILD;
+			progress_vals[2] = newidx->indexId;
+			progress_vals[3] = newidx->amId;
+			pgstat_progress_update_multi_param(4, progress_index, progress_vals);
+		}
 
 		/* Perform concurrent build of new index */
 		index_concurrently_build(newidx->tableId, newidx->indexId);
@@ -4033,14 +4037,16 @@ ReindexRelationConcurrently(Oid relationOid, ReindexParams *params)
 		 * table involved. Don't overwrite CREATE INDEX command.
 		 */
 		if (!(params->options & REINDEXOPT_REPORT_CREATE_PART))
+		{
 			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
 									  newidx->tableId);
 
-		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
-		progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
-		progress_vals[2] = newidx->indexId;
-		progress_vals[3] = newidx->amId;
-		pgstat_progress_update_multi_param(4, progress_index, progress_vals);
+			progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
+			progress_vals[1] = PROGRESS_CREATEIDX_PHASE_VALIDATE_IDXSCAN;
+			progress_vals[2] = newidx->indexId;
+			progress_vals[3] = newidx->amId;
+			pgstat_progress_update_multi_param(4, progress_index, progress_vals);
+		}
 
 		validate_index(newidx->tableId, newidx->indexId, snapshot);
 
-- 
2.30.2

From e25b15173f4ce939efa54426e369b6996129ff59 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                  |   4 +-
 doc/src/sgml/ref/create_index.sgml     |   9 --
 src/backend/commands/indexcmds.c       | 172 +++++++++++++++++--------
 src/test/regress/expected/indexing.out | 127 +++++++++++++++++-
 src/test/regress/sql/indexing.sql      |  26 +++-
 5 files changed, 268 insertions(+), 70 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 38618de01c5..cd72b455447 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4163,9 +4163,7 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      so that they are applied automatically to the entire hierarchy.
      This is very
      convenient, as not only will the existing partitions become indexed, but
-     also any partitions that are created in the future will.  One limitation is
-     that it's not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier when creating such a partitioned index.  To avoid long lock
+     also any partitions that are created in the future will.  To avoid long lock
      times, it is possible to use <command>CREATE INDEX ON ONLY</command>
      the partitioned table; such an index is marked invalid, and the partitions
      do not get the index applied automatically.  The indexes on partitions can
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 40986aa502f..fc8cda655f0 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -692,15 +692,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index b5b860c3abf..cfab45b9992 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -92,6 +92,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(List *colnames);
 static List *ChooseIndexColumnNames(List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(RangeVar *indexRelation, ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -551,7 +556,6 @@ DefineIndex(Oid relationId,
 	bool		amcanorder;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -559,12 +563,10 @@ DefineIndex(Oid relationId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -697,17 +699,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1079,10 +1070,6 @@ DefineIndex(Oid relationId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1147,6 +1134,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1427,12 +1419,15 @@ DefineIndex(Oid relationId,
 								createdConstraintId,
 								is_alter_table, check_rights, check_not_in_use,
 								skip_build, quiet);
+
 					SetUserIdAndSecContext(child_save_userid,
 										   child_save_sec_context);
 				}
 
-				pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
-											 i + 1);
+				/* For concurrent build, this is a catalog-only stage */
+				if (!concurrent)
+					pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+												 i + 1);
 				free_attrmap(attmap);
 			}
 
@@ -1444,46 +1439,39 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 		}
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		return address;
+
+		if (!concurrent)
+		{
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
+
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
-		/* If this is the top-level index, we're done. */
+		/* If this is the top-level index, the command is complete. */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 
@@ -1495,6 +1483,92 @@ DefineIndex(Oid relationId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		DefineIndexConcurrentInternal(relationId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/* finish CIC by building indexes on partitions */
+		ListCell   *lc;
+		List	   *childs;
+		int			npart = 0;
+		MemoryContext cic_context,
+					old_context;
+
+		/*
+		 * Create special memory context for cross-transaction storage.
+		 */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid = IndexGetRelation(indrelid, false);
+
+			if (RELKIND_HAS_STORAGE(get_rel_relkind(indrelid)) &&
+				!get_index_isvalid(indrelid))
+			{
+				rel = table_open(relationId, ShareUpdateExclusiveLock);
+				heaprelid = rel->rd_lockInfo.lockRelId;
+				table_close(rel, ShareUpdateExclusiveLock);
+				SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+				/* Process each partition in a separate transaction */
+				DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+											  heaplocktag, heaprelid);
+
+				PushActiveSnapshot(GetTransactionSnapshot());
+			}
+
+			pgstat_progress_update_param(PROGRESS_CREATEIDX_PARTITIONS_DONE,
+										 ++npart);
+		}
+
+		/* Set all indexes as valid, including the parent */
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+
+			if (get_rel_relkind(indrelid) != RELKIND_PARTITIONED_INDEX)
+				continue;
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid relationId,
+							  Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+					indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1690,10 +1764,6 @@ DefineIndex(Oid relationId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 1bdd430f063..f1beee6d240 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 429120e7104..fb0baedcc28 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.25.1

From 941f7f930fc18563e2da42143015b6573d5447b1 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH] Allow CREATE INDEX CONCURRENTLY on partitioned table

https://www.postgresql.org/message-id/flat/20201031063117.GF3080@telsasoft.com
---
 doc/src/sgml/ddl.sgml                  |   4 +-
 doc/src/sgml/ref/create_index.sgml     |  14 +-
 src/backend/commands/indexcmds.c       | 200 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out | 127 +++++++++++++++-
 src/test/regress/sql/indexing.sql      |  26 +++-
 5 files changed, 297 insertions(+), 74 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 91c036d1cbe..64efdf1e879 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4178,9 +4178,7 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      so that they are applied automatically to the entire hierarchy.
      This is very
      convenient, as not only will the existing partitions become indexed, but
-     also any partitions that are created in the future will.  One limitation is
-     that it's not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier when creating such a partitioned index.  To avoid long lock
+     also any partitions that are created in the future will.  To avoid long lock
      times, it is possible to use <command>CREATE INDEX ON ONLY</command>
      the partitioned table; such an index is marked invalid, and the partitions
      do not get the index applied automatically.  The indexes on partitions can
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 40986aa502f..b05102efdaf 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 3ec8b5cca6c..daba8b67dbe 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -93,6 +93,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(List *colnames);
 static List *ChooseIndexColumnNames(List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(RangeVar *indexRelation, ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -559,7 +564,6 @@ DefineIndex(Oid relationId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -567,12 +571,10 @@ DefineIndex(Oid relationId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -705,17 +707,6 @@ DefineIndex(Oid relationId,
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
 	if (partitioned)
 	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
 		if (stmt->excludeOpNames)
 			ereport(ERROR,
 					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
@@ -1089,10 +1080,6 @@ DefineIndex(Oid relationId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1157,6 +1144,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1494,58 +1486,54 @@ DefineIndex(Oid relationId,
 			 * invalid, this is incorrect, so update our row to invalid too.
 			 */
 			if (invalidate_parent)
-			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
-			}
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 		}
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1556,6 +1544,114 @@ DefineIndex(Oid relationId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(relationId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *partitioned = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				partitioned = lappend_oid(partitioned, indrelid);
+				MemoryContextSwitchTo(old_context);
+				continue;
+			}
+
+			rel = table_open(relationId, ShareUpdateExclusiveLock);
+			heaprelid = rel->rd_lockInfo.lockRelId;
+			table_close(rel, ShareUpdateExclusiveLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			tabrelid = IndexGetRelation(indrelid, false);
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, partitioned)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid relationId,
+							  Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1751,10 +1847,6 @@ DefineIndex(Oid relationId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 1bdd430f063..f1beee6d240 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 429120e7104..fb0baedcc28 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 3f60cbdd12b67115f86854ff60a4009028b8b99f Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzbyj@telsasoft.com>
Date: Sat, 6 Jun 2020 17:42:23 -0500
Subject: [PATCH] Allow CREATE INDEX CONCURRENTLY on partitioned table

https://www.postgresql.org/message-id/flat/20201031063117.GF3080@telsasoft.com
---
 doc/src/sgml/ddl.sgml                  |   4 +-
 doc/src/sgml/ref/create_index.sgml     |  14 +-
 src/backend/commands/indexcmds.c       | 201 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out | 127 +++++++++++++++-
 src/test/regress/sql/indexing.sql      |  26 +++-
 5 files changed, 297 insertions(+), 75 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 58aaa691c6a..afa982154a8 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4161,9 +4161,7 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      so that they are applied automatically to the entire hierarchy.
      This is very
      convenient, as not only will the existing partitions become indexed, but
-     also any partitions that are created in the future will.  One limitation is
-     that it's not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier when creating such a partitioned index.  To avoid long lock
+     also any partitions that are created in the future will.  To avoid long lock
      times, it is possible to use <command>CREATE INDEX ON ONLY</command>
      the partitioned table; such an index is marked invalid, and the partitions
      do not get the index applied automatically.  The indexes on partitions can
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 40986aa502f..b05102efdaf 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index baf3e6e57a5..dfe64052b81 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -92,6 +92,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(List *colnames);
 static List *ChooseIndexColumnNames(List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(RangeVar *indexRelation, ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -555,7 +560,6 @@ DefineIndex(Oid relationId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -563,12 +567,10 @@ DefineIndex(Oid relationId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -699,20 +701,6 @@ DefineIndex(Oid relationId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1102,10 +1090,6 @@ DefineIndex(Oid relationId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1170,6 +1154,11 @@ DefineIndex(Oid relationId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1519,21 +1508,7 @@ DefineIndex(Oid relationId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1545,37 +1520,49 @@ DefineIndex(Oid relationId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1586,6 +1573,114 @@ DefineIndex(Oid relationId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(relationId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				continue;
+			}
+
+			rel = table_open(relationId, ShareUpdateExclusiveLock);
+			heaprelid = rel->rd_lockInfo.lockRelId;
+			table_close(rel, ShareUpdateExclusiveLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			tabrelid = IndexGetRelation(indrelid, false);
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid relationId,
+							  Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1781,10 +1876,6 @@ DefineIndex(Oid relationId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 2be8ffa7ec4..aefa203b14f 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index b69c41832ca..5ddeaf1c613 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 37a13b7fa1c3277b9d038b7a0c75399ff05b28a7 Mon Sep 17 00:00:00 2001
From: Justin Pryzby <pryzby@telsasoft.com>
Date: Mon, 29 Jan 2024 10:41:01 +0300
Subject: [PATCH] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                  |   4 +-
 doc/src/sgml/ref/create_index.sgml     |  14 +-
 src/backend/commands/indexcmds.c       | 200 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out | 127 +++++++++++++++-
 src/test/regress/sql/indexing.sql      |  26 +++-
 5 files changed, 296 insertions(+), 75 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 075ff329912..8ee80c40e3b 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4194,9 +4194,7 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      so that they are applied automatically to the entire hierarchy.
      This is very
      convenient, as not only will the existing partitions become indexed, but
-     also any partitions that are created in the future will.  One limitation is
-     that it's not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier when creating such a partitioned index.  To avoid long lock
+     also any partitions that are created in the future will.  To avoid long lock
      times, it is possible to use <command>CREATE INDEX ON ONLY</command>
      the partitioned table; such an index is marked invalid, and the partitions
      do not get the index applied automatically.  The indexes on partitions can
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 40986aa502f..b05102efdaf 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index ab8b81b3020..65477aeb3a8 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -93,6 +93,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+							  Oid indexRelationId,
+							  IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag,
+							  LockRelId heaprelid);
 static void ReindexIndex(const RangeVar *indexRelation, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -554,7 +559,6 @@ DefineIndex(Oid tableId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -562,12 +566,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -697,20 +699,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1100,10 +1088,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1168,6 +1152,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1516,21 +1505,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1542,37 +1517,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1583,6 +1570,113 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				continue;
+			}
+
+			tabrelid = IndexGetRelation(indrelid, false);
+			rel = table_open(tabrelid, ShareUpdateExclusiveLock);
+			heaprelid = rel->rd_lockInfo.lockRelId;
+			table_close(rel, ShareUpdateExclusiveLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1778,10 +1872,6 @@ DefineIndex(Oid tableId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index 087f955b1e6..5d529454993 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 44f6788915c..96c63615e9a 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 1cb76c8c19b1c0549fbba70febc32017bc04c0a2 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Thu, 23 May 2024 18:13:41 +0100
Subject: [PATCH v2] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                  |   7 +-
 doc/src/sgml/ref/create_index.sgml     |  14 +-
 src/backend/commands/indexcmds.c       | 293 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out | 127 ++++++++++-
 src/test/regress/sql/indexing.sql      |  26 ++-
 5 files changed, 367 insertions(+), 100 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 6aab79e901..f1d4a59a99 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4314,10 +4314,9 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
+     indexed, but any future partitions will be as well.
+     <command>CREATE INDEX ... CONCURRENTLY</command> can incur long lock times
+     on huge partitioned tables, to avoid that you can
      use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 621bc0e253..2366cfd9b5 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 309389e20d..5806eeb8ef 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -95,6 +95,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId *heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -561,7 +566,6 @@ DefineIndex(Oid tableId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -569,12 +573,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -706,20 +708,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1116,10 +1104,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1184,6 +1168,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1533,21 +1522,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1559,37 +1534,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1600,28 +1587,180 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
-	/*
-	 * For a concurrent build, it's important to make the catalog entries
-	 * visible to other transactions before we start to build the index. That
-	 * will prevent them from making incompatible HOT updates.  The new index
-	 * will be marked not indisready and not indisvalid, so that no one else
-	 * tries to either insert into it or use it for queries.
-	 *
-	 * We must commit our current transaction so that the index becomes
-	 * visible; then start another.  Note that all the data structures we just
-	 * built are lost in the commit.  The only data we keep past here are the
-	 * relation IDs.
-	 *
-	 * Before committing, get a session-level lock on the table, to ensure
-	 * that neither it nor the index can be dropped before we finish. This
-	 * cannot block, even if someone else is waiting for access, because we
-	 * already have the same lock within our transaction.
-	 *
-	 * Note: we don't currently bother with a session lock on the index,
-	 * because there are no operations that could change its state while we
-	 * hold lock on the parent table.  This might need to change later.
-	 */
-	LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
+	if (!partitioned)
+	{
+		/*
+		 * For a concurrent build, it's important to make the catalog entries
+		 * visible to other transactions before we start to build the index.
+		 * That will prevent them from making incompatible HOT updates.  The
+		 * new index will be marked not indisready and not indisvalid, so that
+		 * no one else tries to either insert into it or use it for queries.
+		 *
+		 * DefineIndexConcurrentInternal will commit our current transaction
+		 * so that the index becomes visible; then start another.  Note that
+		 * all the data structures we just built are lost in the commit.  The
+		 * only data we keep past here are the relation IDs.
+		 *
+		 * Before committing, get a session-level lock on the table, to ensure
+		 * that neither it nor the index can be dropped before we finish. This
+		 * cannot block, even if someone else is waiting for access, because
+		 * we already have the same lock within our transaction.
+		 *
+		 * Note: we don't currently bother with a session lock on the index,
+		 * because there are no operations that could change its state while
+		 * we hold lock on the parent table.  This might need to change later.
+		 */
+
+		LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
+
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, &heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		ListCell   *lc2;
+		List	   *childs;
+		List	   *part_idxs = NIL;
+		List	   *leaf_idxs = NIL;
+		List	   *leaf_idx_lockids = NIL;
+		List	   *part_idx_lockids = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			char		relkind;
+			LockRelId  *lockrelid;
+			Oid			tabrelid;
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			tabrelid = IndexGetRelation(indrelid, false);
+			rel = table_open(tabrelid, ShareUpdateExclusiveLock);
+			lockrelid = palloc_object(LockRelId);
+			*lockrelid = rel->rd_lockInfo.lockRelId;
+			table_close(rel, ShareUpdateExclusiveLock);
+
+			/*
+			 * Split partitions in 2 lists: partitioned tables that just need
+			 * to be marked as valid and leaf tables that actually need to
+			 * have their indexes built.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				part_idxs = lappend_oid(part_idxs, indrelid);
+				part_idx_lockids = lappend(part_idx_lockids, lockrelid);
+				MemoryContextSwitchTo(old_context);
+			}
+			else
+			{
+				old_context = MemoryContextSwitchTo(cic_context);
+
+				leaf_idxs = lappend_oid(leaf_idxs, indrelid);
+				leaf_idx_lockids = lappend(leaf_idx_lockids, lockrelid);
+
+				MemoryContextSwitchTo(old_context);
+
+			}
+
+			/*
+			 * All partitions, including top-level parent, need to be locked
+			 * for the session before proceeding with any index rebuilds as
+			 * each of those are done in a separate transaction. After each
+			 * leaf index is built, its corresponding table will be unlocked,
+			 * all the partitioned tables will be unlocked at the very end.
+			 */
+
+			LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
+
+		}
+
+		forboth(lc, leaf_idxs, lc2, leaf_idx_lockids)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			LockRelId  *heaprelid = lfirst(lc2);
+			Oid			tabrelid = IndexGetRelation(indrelid, false);
+
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid->dbId, heaprelid->relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, part_idxs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+		}
+
+		foreach(lc, part_idx_lockids)
+		{
+			LockRelId  *lockrelid = lfirst(lc);
+
+			UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
+		}
+
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId *heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
 
 	PopActiveSnapshot();
 	CommitTransactionCommand();
@@ -1789,16 +1928,12 @@ DefineIndex(Oid tableId,
 	 * would be useful.  (Note that our earlier commits did not create reasons
 	 * to replan; so relcache flush on the index itself was sufficient.)
 	 */
-	CacheInvalidateRelcacheByRelid(heaprelid.relId);
+	CacheInvalidateRelcacheByRelid(heaprelid->relId);
 
 	/*
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
-	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
+	UnlockRelationIdForSession(heaprelid, ShareUpdateExclusiveLock);
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index f25723da92..44a6cf39df 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 5f1f4b80c9..38a730d877 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 7687b4c3ba10fc1df5ca4c8f50198dfb269be8bb Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Thu, 23 May 2024 18:13:41 +0100
Subject: [PATCH v3] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                  |  10 +-
 doc/src/sgml/ref/create_index.sgml     |  14 +-
 src/backend/commands/indexcmds.c       | 293 ++++++++++++++++++-------
 src/test/regress/expected/indexing.out | 127 ++++++++++-
 src/test/regress/sql/indexing.sql      |  26 ++-
 5 files changed, 368 insertions(+), 102 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 6aab79e901..904978c6e5 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4314,14 +4314,12 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
-     use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
+     indexed, but any future partitions will be as well. For more control over
+     locking of the partitions you can use <command>CREATE INDEX ON ONLY</command>
+     on the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
-     on each partition using <literal>CONCURRENTLY</literal> and
+     on each partition and
      <firstterm>attached</firstterm> to the partitioned index on the parent
      using <command>ALTER INDEX ... ATTACH PARTITION</command>.  Once indexes for
      all the partitions are attached to the parent index, the parent index will
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 621bc0e253..2366cfd9b5 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 309389e20d..5806eeb8ef 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -95,6 +95,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId *heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -561,7 +566,6 @@ DefineIndex(Oid tableId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -569,12 +573,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -706,20 +708,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1116,10 +1104,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1184,6 +1168,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1533,21 +1522,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1559,37 +1534,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1600,28 +1587,180 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
-	/*
-	 * For a concurrent build, it's important to make the catalog entries
-	 * visible to other transactions before we start to build the index. That
-	 * will prevent them from making incompatible HOT updates.  The new index
-	 * will be marked not indisready and not indisvalid, so that no one else
-	 * tries to either insert into it or use it for queries.
-	 *
-	 * We must commit our current transaction so that the index becomes
-	 * visible; then start another.  Note that all the data structures we just
-	 * built are lost in the commit.  The only data we keep past here are the
-	 * relation IDs.
-	 *
-	 * Before committing, get a session-level lock on the table, to ensure
-	 * that neither it nor the index can be dropped before we finish. This
-	 * cannot block, even if someone else is waiting for access, because we
-	 * already have the same lock within our transaction.
-	 *
-	 * Note: we don't currently bother with a session lock on the index,
-	 * because there are no operations that could change its state while we
-	 * hold lock on the parent table.  This might need to change later.
-	 */
-	LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
+	if (!partitioned)
+	{
+		/*
+		 * For a concurrent build, it's important to make the catalog entries
+		 * visible to other transactions before we start to build the index.
+		 * That will prevent them from making incompatible HOT updates.  The
+		 * new index will be marked not indisready and not indisvalid, so that
+		 * no one else tries to either insert into it or use it for queries.
+		 *
+		 * DefineIndexConcurrentInternal will commit our current transaction
+		 * so that the index becomes visible; then start another.  Note that
+		 * all the data structures we just built are lost in the commit.  The
+		 * only data we keep past here are the relation IDs.
+		 *
+		 * Before committing, get a session-level lock on the table, to ensure
+		 * that neither it nor the index can be dropped before we finish. This
+		 * cannot block, even if someone else is waiting for access, because
+		 * we already have the same lock within our transaction.
+		 *
+		 * Note: we don't currently bother with a session lock on the index,
+		 * because there are no operations that could change its state while
+		 * we hold lock on the parent table.  This might need to change later.
+		 */
+
+		LockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
+
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, &heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		ListCell   *lc2;
+		List	   *childs;
+		List	   *part_idxs = NIL;
+		List	   *leaf_idxs = NIL;
+		List	   *leaf_idx_lockids = NIL;
+		List	   *part_idx_lockids = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			char		relkind;
+			LockRelId  *lockrelid;
+			Oid			tabrelid;
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			tabrelid = IndexGetRelation(indrelid, false);
+			rel = table_open(tabrelid, ShareUpdateExclusiveLock);
+			lockrelid = palloc_object(LockRelId);
+			*lockrelid = rel->rd_lockInfo.lockRelId;
+			table_close(rel, ShareUpdateExclusiveLock);
+
+			/*
+			 * Split partitions in 2 lists: partitioned tables that just need
+			 * to be marked as valid and leaf tables that actually need to
+			 * have their indexes built.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				part_idxs = lappend_oid(part_idxs, indrelid);
+				part_idx_lockids = lappend(part_idx_lockids, lockrelid);
+				MemoryContextSwitchTo(old_context);
+			}
+			else
+			{
+				old_context = MemoryContextSwitchTo(cic_context);
+
+				leaf_idxs = lappend_oid(leaf_idxs, indrelid);
+				leaf_idx_lockids = lappend(leaf_idx_lockids, lockrelid);
+
+				MemoryContextSwitchTo(old_context);
+
+			}
+
+			/*
+			 * All partitions, including top-level parent, need to be locked
+			 * for the session before proceeding with any index rebuilds as
+			 * each of those are done in a separate transaction. After each
+			 * leaf index is built, its corresponding table will be unlocked,
+			 * all the partitioned tables will be unlocked at the very end.
+			 */
+
+			LockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
+
+		}
+
+		forboth(lc, leaf_idxs, lc2, leaf_idx_lockids)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			LockRelId  *heaprelid = lfirst(lc2);
+			Oid			tabrelid = IndexGetRelation(indrelid, false);
+
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid->dbId, heaprelid->relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, part_idxs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+		}
+
+		foreach(lc, part_idx_lockids)
+		{
+			LockRelId  *lockrelid = lfirst(lc);
+
+			UnlockRelationIdForSession(lockrelid, ShareUpdateExclusiveLock);
+		}
+
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId *heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
 
 	PopActiveSnapshot();
 	CommitTransactionCommand();
@@ -1789,16 +1928,12 @@ DefineIndex(Oid tableId,
 	 * would be useful.  (Note that our earlier commits did not create reasons
 	 * to replan; so relcache flush on the index itself was sufficient.)
 	 */
-	CacheInvalidateRelcacheByRelid(heaprelid.relId);
+	CacheInvalidateRelcacheByRelid(heaprelid->relId);
 
 	/*
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
-	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
+	UnlockRelationIdForSession(heaprelid, ShareUpdateExclusiveLock);
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index f25723da92..44a6cf39df 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 5f1f4b80c9..38a730d877 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.34.1

From 884be03aaeabee5c6eeb5a3f639ac9afe712c24b Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Tue, 11 Jun 2024 17:48:08 +0100
Subject: [PATCH v4 2/2] Fix progress report for partitioned REINDEX

---
 src/backend/catalog/index.c      | 11 ++++--
 src/backend/commands/indexcmds.c | 63 +++++++++++++++++++++++++++++---
 src/include/catalog/index.h      |  1 +
 3 files changed, 66 insertions(+), 9 deletions(-)

diff --git a/src/backend/catalog/index.c b/src/backend/catalog/index.c
index 55fdde4b24..c5bc72b350 100644
--- a/src/backend/catalog/index.c
+++ b/src/backend/catalog/index.c
@@ -3559,6 +3559,7 @@ reindex_index(const ReindexStmt *stmt, Oid indexId,
 	volatile bool skipped_constraint = false;
 	PGRUsage	ru0;
 	bool		progress = ((params->options & REINDEXOPT_REPORT_PROGRESS) != 0);
+	bool		partition = ((params->options & REINDEXOPT_PARTITION) != 0);
 	bool		set_tablespace = false;
 
 	pg_rusage_init(&ru0);
@@ -3604,8 +3605,9 @@ reindex_index(const ReindexStmt *stmt, Oid indexId,
 			indexId
 		};
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
-									  heapId);
+		if (!partition)
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+										  heapId);
 		pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
 	}
 
@@ -3845,8 +3847,11 @@ reindex_index(const ReindexStmt *stmt, Oid indexId,
 	index_close(iRel, NoLock);
 	table_close(heapRelation, NoLock);
 
-	if (progress)
+	if (progress && !partition)
+	{
+		/* progress for partitions is tracked in the caller */
 		pgstat_progress_end_command();
+	}
 }
 
 /*
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 5da2df2d3b..17b30ad6aa 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -118,6 +118,7 @@ static bool ReindexRelationConcurrently(const ReindexStmt *stmt,
 										const ReindexParams *params);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
+static inline void progress_index_partition_done(void);
 
 /*
  * callback argument type for RangeVarCallbackForReindexIndex()
@@ -1550,7 +1551,7 @@ DefineIndex(Oid tableId,
 				 * Update progress for an intermediate partitioned index
 				 * itself
 				 */
-				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				progress_index_partition_done();
 			}
 
 			return address;
@@ -1577,7 +1578,7 @@ DefineIndex(Oid tableId,
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 		else if (!concurrent)
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			progress_index_partition_done();
 
 		return address;
 	}
@@ -1682,7 +1683,7 @@ DefineIndex(Oid tableId,
 										  heaplocktag, heaprelid);
 
 			PushActiveSnapshot(GetTransactionSnapshot());
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			progress_index_partition_done();
 		}
 
 		/* Set as valid all partitioned indexes, including the parent */
@@ -3327,6 +3328,14 @@ ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *param
 	ListCell   *lc;
 	ErrorContextCallback errcallback;
 	ReindexErrorInfo errinfo;
+	ReindexParams newparams;
+	int			progress_params[3] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_PARTITIONS_TOTAL
+	};
+	int64		progress_values[3];
+	Oid			heapId = relid;
 
 	Assert(RELKIND_HAS_PARTITIONS(relkind));
 
@@ -3388,11 +3397,28 @@ ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *param
 		MemoryContextSwitchTo(old_context);
 	}
 
+	if (relkind == RELKIND_PARTITIONED_INDEX)
+	{
+		heapId = IndexGetRelation(relid, true);
+	}
+
+	progress_values[0] = (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
+		PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY :
+		PROGRESS_CREATEIDX_COMMAND_REINDEX;
+	progress_values[1] = 0;
+	progress_values[2] = list_length(partitions);
+	pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, heapId);
+	pgstat_progress_update_multi_param(3, progress_params, progress_values);
+
 	/*
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(stmt, partitions, params);
+	newparams = *params;
+	newparams.options |= REINDEXOPT_PARTITION;
+	ReindexMultipleInternal(stmt, partitions, &newparams);
+
+	pgstat_progress_end_command();
 
 	/*
 	 * Clean up working storage --- note we must do this after
@@ -3413,6 +3439,7 @@ static void
 ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
 {
 	ListCell   *l;
+	bool		partitions = ((params->options & REINDEXOPT_PARTITION) != 0);
 
 	PopActiveSnapshot();
 	CommitTransactionCommand();
@@ -3506,6 +3533,9 @@ ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const Reind
 		}
 
 		CommitTransactionCommand();
+
+		if (partitions)
+			progress_index_partition_done();
 	}
 
 	StartTransactionCommand();
@@ -3558,6 +3588,7 @@ ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const Rein
 	char	   *relationName = NULL;
 	char	   *relationNamespace = NULL;
 	PGRUsage	ru0;
+	bool		partition = ((params->options & REINDEXOPT_PARTITION) != 0);
 	const int	progress_index[] = {
 		PROGRESS_CREATEIDX_COMMAND,
 		PROGRESS_CREATEIDX_PHASE,
@@ -3901,7 +3932,8 @@ ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const Rein
 		if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
 			elog(ERROR, "cannot reindex a temporary table concurrently");
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
+		if (!partition)
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
 
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = 0;	/* initializing */
@@ -4375,7 +4407,8 @@ ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const Rein
 
 	MemoryContextDelete(private_context);
 
-	pgstat_progress_end_command();
+	if (!partition)
+		pgstat_progress_end_command();
 
 	return true;
 }
@@ -4565,3 +4598,21 @@ set_indexsafe_procflags(void)
 	ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
 	LWLockRelease(ProcArrayLock);
 }
+
+static inline void
+progress_index_partition_done(void)
+{
+	int			nparam = 6;
+	const int	progress_idx[] = {
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_SUBPHASE,
+		PROGRESS_CREATEIDX_TUPLES_TOTAL,
+		PROGRESS_CREATEIDX_TUPLES_DONE
+	};
+	const int64 progress_vals[] = {0, 0, 0, 0, 0, 0};
+
+	pgstat_progress_update_multi_param(nparam, progress_idx, progress_vals);
+	pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+}
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index 7d434f8e65..ccba65fbbf 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_PARTITION 0x10 /* reindexing is done as part of partitioned table/index reindex */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.43.0

From 45576292012468f2d3deea0e568ee74891cb73b1 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Thu, 23 May 2024 18:13:41 +0100
Subject: [PATCH v4 1/2] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                  |  10 +-
 doc/src/sgml/ref/create_index.sgml     |  14 +-
 src/backend/commands/indexcmds.c       | 224 +++++++++++++++++++------
 src/test/regress/expected/indexing.out | 127 +++++++++++++-
 src/test/regress/sql/indexing.sql      |  26 ++-
 5 files changed, 323 insertions(+), 78 deletions(-)

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 6aab79e901..904978c6e5 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4314,14 +4314,12 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
-     use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
+     indexed, but any future partitions will be as well. For more control over
+     locking of the partitions you can use <command>CREATE INDEX ON ONLY</command>
+     on the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
-     on each partition using <literal>CONCURRENTLY</literal> and
+     on each partition and
      <firstterm>attached</firstterm> to the partitioned index on the parent
      using <command>ALTER INDEX ... ATTACH PARTITION</command>.  Once indexes for
      all the partitions are attached to the parent index, the parent index will
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 621bc0e253..2366cfd9b5 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 309389e20d..5da2df2d3b 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -95,6 +95,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -561,7 +566,6 @@ DefineIndex(Oid tableId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -569,12 +573,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -706,20 +708,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1116,10 +1104,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1184,6 +1168,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1533,21 +1522,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1559,37 +1534,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1600,6 +1587,137 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+				continue;
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				continue;
+			}
+
+			/*
+			 * Partition could have been dropped, since we looked it up. In
+			 * this case consider it done and go to the next one.
+			 */
+			tabrelid = IndexGetRelation(indrelid, true);
+			if (!tabrelid)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+			rel = try_table_open(tabrelid, ShareUpdateExclusiveLock);
+			if (!rel)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+
+			heaprelid = rel->rd_lockInfo.lockRelId;
+
+			/*
+			 * Close the table but retain the lock, that should be extended to
+			 * session level in DefineIndexConcurrentInternal.
+			 */
+			table_close(rel, NoLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Relation	indrel = try_index_open(indrelid, ShareUpdateExclusiveLock);
+
+			if (!indrel)
+				continue;
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+			index_close(indrel, ShareUpdateExclusiveLock);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1795,10 +1913,6 @@ DefineIndex(Oid tableId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index f25723da92..44a6cf39df 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 5f1f4b80c9..38a730d877 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.43.0

From 45f2ec9ee57a5337b77b66db3c8c5092f305a176 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Tue, 11 Jun 2024 17:48:08 +0100
Subject: [PATCH v5 2/2] Fix progress report for partitioned REINDEX

---
 src/backend/catalog/index.c      | 11 ++++--
 src/backend/commands/indexcmds.c | 63 +++++++++++++++++++++++++++++---
 src/include/catalog/index.h      |  1 +
 3 files changed, 66 insertions(+), 9 deletions(-)

diff --git a/src/backend/catalog/index.c b/src/backend/catalog/index.c
index 55fdde4b24..c5bc72b350 100644
--- a/src/backend/catalog/index.c
+++ b/src/backend/catalog/index.c
@@ -3559,6 +3559,7 @@ reindex_index(const ReindexStmt *stmt, Oid indexId,
 	volatile bool skipped_constraint = false;
 	PGRUsage	ru0;
 	bool		progress = ((params->options & REINDEXOPT_REPORT_PROGRESS) != 0);
+	bool		partition = ((params->options & REINDEXOPT_PARTITION) != 0);
 	bool		set_tablespace = false;
 
 	pg_rusage_init(&ru0);
@@ -3604,8 +3605,9 @@ reindex_index(const ReindexStmt *stmt, Oid indexId,
 			indexId
 		};
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
-									  heapId);
+		if (!partition)
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX,
+										  heapId);
 		pgstat_progress_update_multi_param(2, progress_cols, progress_vals);
 	}
 
@@ -3845,8 +3847,11 @@ reindex_index(const ReindexStmt *stmt, Oid indexId,
 	index_close(iRel, NoLock);
 	table_close(heapRelation, NoLock);
 
-	if (progress)
+	if (progress && !partition)
+	{
+		/* progress for partitions is tracked in the caller */
 		pgstat_progress_end_command();
+	}
 }
 
 /*
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index dcb4ea89e9..6abe1f017c 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -118,6 +118,7 @@ static bool ReindexRelationConcurrently(const ReindexStmt *stmt,
 										const ReindexParams *params);
 static void update_relispartition(Oid relationId, bool newval);
 static inline void set_indexsafe_procflags(void);
+static inline void progress_index_partition_done(void);
 
 /*
  * callback argument type for RangeVarCallbackForReindexIndex()
@@ -1550,7 +1551,7 @@ DefineIndex(Oid tableId,
 				 * Update progress for an intermediate partitioned index
 				 * itself
 				 */
-				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				progress_index_partition_done();
 			}
 
 			return address;
@@ -1577,7 +1578,7 @@ DefineIndex(Oid tableId,
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
 		else if (!concurrent)
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			progress_index_partition_done();
 
 		return address;
 	}
@@ -1686,7 +1687,7 @@ DefineIndex(Oid tableId,
 										  heaplocktag, heaprelid);
 
 			PushActiveSnapshot(GetTransactionSnapshot());
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			progress_index_partition_done();
 		}
 
 		/* Set as valid all partitioned indexes, including the parent */
@@ -3331,6 +3332,14 @@ ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *param
 	ListCell   *lc;
 	ErrorContextCallback errcallback;
 	ReindexErrorInfo errinfo;
+	ReindexParams newparams;
+	int			progress_params[3] = {
+		PROGRESS_CREATEIDX_COMMAND,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_PARTITIONS_TOTAL
+	};
+	int64		progress_values[3];
+	Oid			heapId = relid;
 
 	Assert(RELKIND_HAS_PARTITIONS(relkind));
 
@@ -3392,11 +3401,28 @@ ReindexPartitions(const ReindexStmt *stmt, Oid relid, const ReindexParams *param
 		MemoryContextSwitchTo(old_context);
 	}
 
+	if (relkind == RELKIND_PARTITIONED_INDEX)
+	{
+		heapId = IndexGetRelation(relid, true);
+	}
+
+	progress_values[0] = (params->options & REINDEXOPT_CONCURRENTLY) != 0 ?
+		PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY :
+		PROGRESS_CREATEIDX_COMMAND_REINDEX;
+	progress_values[1] = 0;
+	progress_values[2] = list_length(partitions);
+	pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, heapId);
+	pgstat_progress_update_multi_param(3, progress_params, progress_values);
+
 	/*
 	 * Process each partition listed in a separate transaction.  Note that
 	 * this commits and then starts a new transaction immediately.
 	 */
-	ReindexMultipleInternal(stmt, partitions, params);
+	newparams = *params;
+	newparams.options |= REINDEXOPT_PARTITION;
+	ReindexMultipleInternal(stmt, partitions, &newparams);
+
+	pgstat_progress_end_command();
 
 	/*
 	 * Clean up working storage --- note we must do this after
@@ -3417,6 +3443,7 @@ static void
 ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const ReindexParams *params)
 {
 	ListCell   *l;
+	bool		partitions = ((params->options & REINDEXOPT_PARTITION) != 0);
 
 	PopActiveSnapshot();
 	CommitTransactionCommand();
@@ -3510,6 +3537,9 @@ ReindexMultipleInternal(const ReindexStmt *stmt, const List *relids, const Reind
 		}
 
 		CommitTransactionCommand();
+
+		if (partitions)
+			progress_index_partition_done();
 	}
 
 	StartTransactionCommand();
@@ -3562,6 +3592,7 @@ ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const Rein
 	char	   *relationName = NULL;
 	char	   *relationNamespace = NULL;
 	PGRUsage	ru0;
+	bool		partition = ((params->options & REINDEXOPT_PARTITION) != 0);
 	const int	progress_index[] = {
 		PROGRESS_CREATEIDX_COMMAND,
 		PROGRESS_CREATEIDX_PHASE,
@@ -3905,7 +3936,8 @@ ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const Rein
 		if (indexRel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
 			elog(ERROR, "cannot reindex a temporary table concurrently");
 
-		pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
+		if (!partition)
+			pgstat_progress_start_command(PROGRESS_COMMAND_CREATE_INDEX, idx->tableId);
 
 		progress_vals[0] = PROGRESS_CREATEIDX_COMMAND_REINDEX_CONCURRENTLY;
 		progress_vals[1] = 0;	/* initializing */
@@ -4379,7 +4411,8 @@ ReindexRelationConcurrently(const ReindexStmt *stmt, Oid relationOid, const Rein
 
 	MemoryContextDelete(private_context);
 
-	pgstat_progress_end_command();
+	if (!partition)
+		pgstat_progress_end_command();
 
 	return true;
 }
@@ -4569,3 +4602,21 @@ set_indexsafe_procflags(void)
 	ProcGlobal->statusFlags[MyProc->pgxactoff] = MyProc->statusFlags;
 	LWLockRelease(ProcArrayLock);
 }
+
+static inline void
+progress_index_partition_done(void)
+{
+	int			nparam = 6;
+	const int	progress_idx[] = {
+		PROGRESS_CREATEIDX_INDEX_OID,
+		PROGRESS_CREATEIDX_ACCESS_METHOD_OID,
+		PROGRESS_CREATEIDX_PHASE,
+		PROGRESS_CREATEIDX_SUBPHASE,
+		PROGRESS_CREATEIDX_TUPLES_TOTAL,
+		PROGRESS_CREATEIDX_TUPLES_DONE
+	};
+	const int64 progress_vals[] = {0, 0, 0, 0, 0, 0};
+
+	pgstat_progress_update_multi_param(nparam, progress_idx, progress_vals);
+	pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+}
diff --git a/src/include/catalog/index.h b/src/include/catalog/index.h
index 7d434f8e65..ccba65fbbf 100644
--- a/src/include/catalog/index.h
+++ b/src/include/catalog/index.h
@@ -42,6 +42,7 @@ typedef struct ReindexParams
 #define REINDEXOPT_REPORT_PROGRESS 0x02 /* report pgstat progress */
 #define REINDEXOPT_MISSING_OK 	0x04	/* skip missing relations */
 #define REINDEXOPT_CONCURRENTLY	0x08	/* concurrent mode */
+#define REINDEXOPT_PARTITION 0x10 /* reindexing is done as part of partitioned table/index reindex */
 
 /* state info for validate_index bulkdelete callback */
 typedef struct ValidateIndexState
-- 
2.43.0

From acf5cf5d4a984c0f8635a25e03c23409601c0c93 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Thu, 23 May 2024 18:13:41 +0100
Subject: [PATCH v5 1/2] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                         |  10 +-
 doc/src/sgml/ref/create_index.sgml            |  14 +-
 src/backend/commands/indexcmds.c              | 228 +++++++++++++-----
 .../isolation/expected/partitioned-cic.out    | 135 +++++++++++
 src/test/isolation/isolation_schedule         |   1 +
 src/test/isolation/specs/partitioned-cic.spec |  57 +++++
 src/test/regress/expected/indexing.out        | 127 +++++++++-
 src/test/regress/sql/indexing.sql             |  26 +-
 8 files changed, 520 insertions(+), 78 deletions(-)
 create mode 100644 src/test/isolation/expected/partitioned-cic.out
 create mode 100644 src/test/isolation/specs/partitioned-cic.spec

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 6aab79e901..904978c6e5 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4314,14 +4314,12 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
-     use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
+     indexed, but any future partitions will be as well. For more control over
+     locking of the partitions you can use <command>CREATE INDEX ON ONLY</command>
+     on the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
-     on each partition using <literal>CONCURRENTLY</literal> and
+     on each partition and
      <firstterm>attached</firstterm> to the partitioned index on the parent
      using <command>ALTER INDEX ... ATTACH PARTITION</command>.  Once indexes for
      all the partitions are attached to the parent index, the parent index will
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 621bc0e253..2366cfd9b5 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 309389e20d..dcb4ea89e9 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -95,6 +95,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -561,7 +566,6 @@ DefineIndex(Oid tableId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -569,12 +573,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -706,20 +708,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1116,10 +1104,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1184,6 +1168,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1533,21 +1522,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1559,37 +1534,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1600,6 +1587,141 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Partition could have been dropped, since we looked it up. In
+			 * this case consider it done and go to the next one.
+			 */
+			tabrelid = IndexGetRelation(indrelid, true);
+			if (!tabrelid)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+			rel = try_table_open(tabrelid, ShareUpdateExclusiveLock);
+			if (!rel)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+			{
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			heaprelid = rel->rd_lockInfo.lockRelId;
+
+			/*
+			 * Close the table but retain the lock, that should be extended to
+			 * session level in DefineIndexConcurrentInternal.
+			 */
+			table_close(rel, NoLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Relation	indrel = try_index_open(indrelid, ShareUpdateExclusiveLock);
+
+			if (!indrel)
+				continue;
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+			index_close(indrel, ShareUpdateExclusiveLock);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1795,10 +1917,6 @@ DefineIndex(Oid tableId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/isolation/expected/partitioned-cic.out b/src/test/isolation/expected/partitioned-cic.out
new file mode 100644
index 0000000000..b66acc6f6a
--- /dev/null
+++ b/src/test/isolation/expected/partitioned-cic.out
@@ -0,0 +1,135 @@
+Parsed test spec with 3 sessions
+
+starting permutation: lock_p1 cic insert drop2 commit chk_content
+step lock_p1: lock cictab_part_1 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step drop2: DROP TABLE cictab_part_2;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
+
+starting permutation: lock_p2 cic insert drop1 commit chk_content
+step lock_p2: lock cictab_part_2 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step drop1: DROP TABLE cictab_part_1;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+
+starting permutation: lock_p1 cic insert detach2 commit chk_content chk_content_part2
+step lock_p1: lock cictab_part_1 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step detach2: ALTER TABLE cictab DETACH PARTITION cictab_part_2;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
+step chk_content_part2: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_2 where i > 0;
+  select * from cictab_part_2 where i > 0;
+
+QUERY PLAN                                           
+-----------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2
+  Index Cond: (i > 0)                                
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+
+starting permutation: lock_p2 cic insert detach1 commit chk_content chk_content_part1
+step lock_p2: lock cictab_part_2 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step detach1: ALTER TABLE cictab DETACH PARTITION cictab_part_1;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+step chk_content_part1: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_1 where i > 0;
+  select * from cictab_part_1 where i > 0;
+
+QUERY PLAN                                           
+-----------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1
+  Index Cond: (i > 0)                                
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule
index 0342eb39e4..57b7948687 100644
--- a/src/test/isolation/isolation_schedule
+++ b/src/test/isolation/isolation_schedule
@@ -114,3 +114,4 @@ test: serializable-parallel-2
 test: serializable-parallel-3
 test: matview-write-skew
 test: lock-nowait
+test: partitioned-cic
diff --git a/src/test/isolation/specs/partitioned-cic.spec b/src/test/isolation/specs/partitioned-cic.spec
new file mode 100644
index 0000000000..95f0bb2b47
--- /dev/null
+++ b/src/test/isolation/specs/partitioned-cic.spec
@@ -0,0 +1,57 @@
+# Test the ability to drop/detach partitions while CREATE INDEX CONCURRENTLY is running.
+# To achieve this, start a transaction that will pause CIC in progress by
+# locking a partition in row exclusive mode, giving us a change to drop/detach another partition.
+# Dropping/detaching is tested for each partition to test two scenarios:
+# when the partition has already been indexed and when it's yet to be indexed.
+
+setup {
+  create table cictab(i int, j int) partition by range(i);
+  create table cictab_part_1 partition of cictab for values from (0) to (10);
+  create table cictab_part_2 partition of cictab for values from (10) to (20);
+
+  insert into cictab values (1, 0), (11, 0);
+}
+
+teardown {
+    drop table if exists cictab_part_1;
+    drop table if exists cictab_part_2;
+    drop table cictab;
+}
+
+session s1
+setup {BEGIN;}
+step lock_p1 { lock cictab_part_1 in row exclusive mode; }
+step lock_p2 { lock cictab_part_2 in row exclusive mode; }
+step commit { COMMIT; }
+
+session s2
+step cic { CREATE INDEX CONCURRENTLY ON cictab(i); }
+step chk_content {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+}
+
+step chk_content_part1 {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_1 where i > 0;
+  select * from cictab_part_1 where i > 0;
+}
+
+step chk_content_part2 {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_2 where i > 0;
+  select * from cictab_part_2 where i > 0;
+}
+
+session s3
+step detach1 { ALTER TABLE cictab DETACH PARTITION cictab_part_1; }
+step detach2 { ALTER TABLE cictab DETACH PARTITION cictab_part_2; }
+step drop1 { DROP TABLE cictab_part_1; }
+step drop2 { DROP TABLE cictab_part_2; }
+step insert { insert into cictab values (1, 1), (11, 1); }
+
+permutation lock_p1 cic insert drop2 commit chk_content
+permutation lock_p2 cic insert drop1 commit chk_content
+permutation lock_p1 cic insert detach2 commit chk_content chk_content_part2
+permutation lock_p2 cic insert detach1 commit chk_content chk_content_part1
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index f25723da92..44a6cf39df 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 5f1f4b80c9..38a730d877 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.43.0

From acf5cf5d4a984c0f8635a25e03c23409601c0c93 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Thu, 23 May 2024 18:13:41 +0100
Subject: [PATCH v5 1/2] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                         |  10 +-
 doc/src/sgml/ref/create_index.sgml            |  14 +-
 src/backend/commands/indexcmds.c              | 228 +++++++++++++-----
 .../isolation/expected/partitioned-cic.out    | 135 +++++++++++
 src/test/isolation/isolation_schedule         |   1 +
 src/test/isolation/specs/partitioned-cic.spec |  57 +++++
 src/test/regress/expected/indexing.out        | 127 +++++++++-
 src/test/regress/sql/indexing.sql             |  26 +-
 8 files changed, 520 insertions(+), 78 deletions(-)
 create mode 100644 src/test/isolation/expected/partitioned-cic.out
 create mode 100644 src/test/isolation/specs/partitioned-cic.spec

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 6aab79e901..904978c6e5 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4314,14 +4314,12 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
-     use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
+     indexed, but any future partitions will be as well. For more control over
+     locking of the partitions you can use <command>CREATE INDEX ON ONLY</command>
+     on the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
-     on each partition using <literal>CONCURRENTLY</literal> and
+     on each partition and
      <firstterm>attached</firstterm> to the partitioned index on the parent
      using <command>ALTER INDEX ... ATTACH PARTITION</command>.  Once indexes for
      all the partitions are attached to the parent index, the parent index will
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index 621bc0e253..2366cfd9b5 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -645,7 +645,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -692,15 +695,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 309389e20d..dcb4ea89e9 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -95,6 +95,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -561,7 +566,6 @@ DefineIndex(Oid tableId,
 	bool		amissummarizing;
 	amoptions_function amoptions;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -569,12 +573,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -706,20 +708,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1116,10 +1104,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1184,6 +1168,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1533,21 +1522,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1559,37 +1534,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1600,6 +1587,141 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Partition could have been dropped, since we looked it up. In
+			 * this case consider it done and go to the next one.
+			 */
+			tabrelid = IndexGetRelation(indrelid, true);
+			if (!tabrelid)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+			rel = try_table_open(tabrelid, ShareUpdateExclusiveLock);
+			if (!rel)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+			{
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			heaprelid = rel->rd_lockInfo.lockRelId;
+
+			/*
+			 * Close the table but retain the lock, that should be extended to
+			 * session level in DefineIndexConcurrentInternal.
+			 */
+			table_close(rel, NoLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Relation	indrel = try_index_open(indrelid, ShareUpdateExclusiveLock);
+
+			if (!indrel)
+				continue;
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+			index_close(indrel, ShareUpdateExclusiveLock);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1795,10 +1917,6 @@ DefineIndex(Oid tableId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/isolation/expected/partitioned-cic.out b/src/test/isolation/expected/partitioned-cic.out
new file mode 100644
index 0000000000..b66acc6f6a
--- /dev/null
+++ b/src/test/isolation/expected/partitioned-cic.out
@@ -0,0 +1,135 @@
+Parsed test spec with 3 sessions
+
+starting permutation: lock_p1 cic insert drop2 commit chk_content
+step lock_p1: lock cictab_part_1 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step drop2: DROP TABLE cictab_part_2;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
+
+starting permutation: lock_p2 cic insert drop1 commit chk_content
+step lock_p2: lock cictab_part_2 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step drop1: DROP TABLE cictab_part_1;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+
+starting permutation: lock_p1 cic insert detach2 commit chk_content chk_content_part2
+step lock_p1: lock cictab_part_1 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step detach2: ALTER TABLE cictab DETACH PARTITION cictab_part_2;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
+step chk_content_part2: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_2 where i > 0;
+  select * from cictab_part_2 where i > 0;
+
+QUERY PLAN                                           
+-----------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2
+  Index Cond: (i > 0)                                
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+
+starting permutation: lock_p2 cic insert detach1 commit chk_content chk_content_part1
+step lock_p2: lock cictab_part_2 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step detach1: ALTER TABLE cictab DETACH PARTITION cictab_part_1;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+step chk_content_part1: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_1 where i > 0;
+  select * from cictab_part_1 where i > 0;
+
+QUERY PLAN                                           
+-----------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1
+  Index Cond: (i > 0)                                
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule
index 0342eb39e4..57b7948687 100644
--- a/src/test/isolation/isolation_schedule
+++ b/src/test/isolation/isolation_schedule
@@ -114,3 +114,4 @@ test: serializable-parallel-2
 test: serializable-parallel-3
 test: matview-write-skew
 test: lock-nowait
+test: partitioned-cic
diff --git a/src/test/isolation/specs/partitioned-cic.spec b/src/test/isolation/specs/partitioned-cic.spec
new file mode 100644
index 0000000000..95f0bb2b47
--- /dev/null
+++ b/src/test/isolation/specs/partitioned-cic.spec
@@ -0,0 +1,57 @@
+# Test the ability to drop/detach partitions while CREATE INDEX CONCURRENTLY is running.
+# To achieve this, start a transaction that will pause CIC in progress by
+# locking a partition in row exclusive mode, giving us a change to drop/detach another partition.
+# Dropping/detaching is tested for each partition to test two scenarios:
+# when the partition has already been indexed and when it's yet to be indexed.
+
+setup {
+  create table cictab(i int, j int) partition by range(i);
+  create table cictab_part_1 partition of cictab for values from (0) to (10);
+  create table cictab_part_2 partition of cictab for values from (10) to (20);
+
+  insert into cictab values (1, 0), (11, 0);
+}
+
+teardown {
+    drop table if exists cictab_part_1;
+    drop table if exists cictab_part_2;
+    drop table cictab;
+}
+
+session s1
+setup {BEGIN;}
+step lock_p1 { lock cictab_part_1 in row exclusive mode; }
+step lock_p2 { lock cictab_part_2 in row exclusive mode; }
+step commit { COMMIT; }
+
+session s2
+step cic { CREATE INDEX CONCURRENTLY ON cictab(i); }
+step chk_content {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+}
+
+step chk_content_part1 {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_1 where i > 0;
+  select * from cictab_part_1 where i > 0;
+}
+
+step chk_content_part2 {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_2 where i > 0;
+  select * from cictab_part_2 where i > 0;
+}
+
+session s3
+step detach1 { ALTER TABLE cictab DETACH PARTITION cictab_part_1; }
+step detach2 { ALTER TABLE cictab DETACH PARTITION cictab_part_2; }
+step drop1 { DROP TABLE cictab_part_1; }
+step drop2 { DROP TABLE cictab_part_2; }
+step insert { insert into cictab values (1, 1), (11, 1); }
+
+permutation lock_p1 cic insert drop2 commit chk_content
+permutation lock_p2 cic insert drop1 commit chk_content
+permutation lock_p1 cic insert detach2 commit chk_content chk_content_part2
+permutation lock_p2 cic insert detach1 commit chk_content chk_content_part1
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index f25723da92..44a6cf39df 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,11 +50,130 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+    "idxpart_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+    "idxpart1_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+    "idxpart11_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+    "idxpart111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+    "idxpart1111_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+    "idxpart2_a_idx1" UNIQUE, btree (a) INVALID
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+    "idxpart3_a_idx1" UNIQUE, btree (a) INVALID
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+    "idxpart31_a_idx1" UNIQUE, btree (a) INVALID
+
 drop table idxpart;
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index 5f1f4b80c9..38a730d877 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,10 +29,30 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
 drop table idxpart;
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
-- 
2.43.0

From a29654340a50f29cb31a665899bf4635fee295b8 Mon Sep 17 00:00:00 2001
From: Ilya Gladyshev <ilya.v.gladyshev@gmail.com>
Date: Tue, 1 Jul 2025 08:59:02 +0300
Subject: [PATCH] Allow CREATE INDEX CONCURRENTLY on partitioned table

---
 doc/src/sgml/ddl.sgml                         |  10 +-
 doc/src/sgml/ref/create_index.sgml            |  14 +-
 src/backend/commands/indexcmds.c              | 231 +++++++++++++-----
 .../isolation/expected/partitioned-cic.out    | 135 ++++++++++
 src/test/isolation/isolation_schedule         |   1 +
 src/test/isolation/specs/partitioned-cic.spec |  57 +++++
 src/test/regress/expected/indexing.out        | 132 +++++++++-
 src/test/regress/sql/indexing.sql             |  42 +++-
 8 files changed, 543 insertions(+), 79 deletions(-)
 create mode 100644 src/test/isolation/expected/partitioned-cic.out
 create mode 100644 src/test/isolation/specs/partitioned-cic.spec

diff --git a/doc/src/sgml/ddl.sgml b/doc/src/sgml/ddl.sgml
index 65bc070d2e5..85c53f93cd5 100644
--- a/doc/src/sgml/ddl.sgml
+++ b/doc/src/sgml/ddl.sgml
@@ -4418,14 +4418,12 @@ ALTER TABLE measurement ATTACH PARTITION measurement_y2008m02
      As mentioned earlier, it is possible to create indexes on partitioned
      tables so that they are applied automatically to the entire hierarchy.
      This can be very convenient as not only will all existing partitions be
-     indexed, but any future partitions will be as well.  However, one
-     limitation when creating new indexes on partitioned tables is that it
-     is not possible to use the <literal>CONCURRENTLY</literal>
-     qualifier, which could lead to long lock times.  To avoid this, you can
-     use <command>CREATE INDEX ON ONLY</command> the partitioned table, which
+     indexed, but any future partitions will be as well. For more control over
+     locking of the partitions you can use <command>CREATE INDEX ON ONLY</command>
+     on the partitioned table, which
      creates the new index marked as invalid, preventing automatic application
      to existing partitions.  Instead, indexes can then be created individually
-     on each partition using <literal>CONCURRENTLY</literal> and
+     on each partition and
      <firstterm>attached</firstterm> to the partitioned index on the parent
      using <command>ALTER INDEX ... ATTACH PARTITION</command>.  Once indexes for
      all the partitions are attached to the parent index, the parent index will
diff --git a/doc/src/sgml/ref/create_index.sgml b/doc/src/sgml/ref/create_index.sgml
index b9c679c41e8..253e62e7348 100644
--- a/doc/src/sgml/ref/create_index.sgml
+++ b/doc/src/sgml/ref/create_index.sgml
@@ -651,7 +651,10 @@ CREATE [ UNIQUE ] INDEX [ CONCURRENTLY ] [ [ IF NOT EXISTS ] <replaceable class=
    <para>
     If a problem arises while scanning the table, such as a deadlock or a
     uniqueness violation in a unique index, the <command>CREATE INDEX</command>
-    command will fail but leave behind an <quote>invalid</quote> index. This index
+    command will fail but leave behind an <quote>invalid</quote> index.
+    If this happens while build an index concurrently on a partitioned
+    table, the command can also leave behind <quote>valid</quote> or
+    <quote>invalid</quote> indexes on table partitions.  The invalid index
     will be ignored for querying purposes because it might be incomplete;
     however it will still consume update overhead. The <application>psql</application>
     <command>\d</command> command will report such an index as <literal>INVALID</literal>:
@@ -698,15 +701,6 @@ Indexes:
     cannot.
    </para>
 
-   <para>
-    Concurrent builds for indexes on partitioned tables are currently not
-    supported.  However, you may concurrently build the index on each
-    partition individually and then finally create the partitioned index
-    non-concurrently in order to reduce the time where writes to the
-    partitioned table will be locked out.  In this case, building the
-    partitioned index is a metadata only operation.
-   </para>
-
   </refsect2>
  </refsect1>
 
diff --git a/src/backend/commands/indexcmds.c b/src/backend/commands/indexcmds.c
index 6f753ab6d7a..c39b3b18ea8 100644
--- a/src/backend/commands/indexcmds.c
+++ b/src/backend/commands/indexcmds.c
@@ -98,6 +98,11 @@ static char *ChooseIndexName(const char *tabname, Oid namespaceId,
 							 bool primary, bool isconstraint);
 static char *ChooseIndexNameAddition(const List *colnames);
 static List *ChooseIndexColumnNames(const List *indexElems);
+static void DefineIndexConcurrentInternal(Oid relationId,
+										  Oid indexRelationId,
+										  IndexInfo *indexInfo,
+										  LOCKTAG heaplocktag,
+										  LockRelId heaprelid);
 static void ReindexIndex(const ReindexStmt *stmt, const ReindexParams *params,
 						 bool isTopLevel);
 static void RangeVarCallbackForReindexIndex(const RangeVar *relation,
@@ -573,7 +578,6 @@ DefineIndex(Oid tableId,
 	amoptions_function amoptions;
 	bool		exclusion;
 	bool		partitioned;
-	bool		safe_index;
 	Datum		reloptions;
 	int16	   *coloptions;
 	IndexInfo  *indexInfo;
@@ -581,12 +585,10 @@ DefineIndex(Oid tableId,
 	bits16		constr_flags;
 	int			numberOfAttributes;
 	int			numberOfKeyAttributes;
-	TransactionId limitXmin;
 	ObjectAddress address;
 	LockRelId	heaprelid;
 	LOCKTAG		heaplocktag;
 	LOCKMODE	lockmode;
-	Snapshot	snapshot;
 	Oid			root_save_userid;
 	int			root_save_sec_context;
 	int			root_save_nestlevel;
@@ -724,20 +726,6 @@ DefineIndex(Oid tableId,
 	 * partition.
 	 */
 	partitioned = rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE;
-	if (partitioned)
-	{
-		/*
-		 * Note: we check 'stmt->concurrent' rather than 'concurrent', so that
-		 * the error is thrown also for temporary tables.  Seems better to be
-		 * consistent, even though we could do it on temporary table because
-		 * we're not actually doing it concurrently.
-		 */
-		if (stmt->concurrent)
-			ereport(ERROR,
-					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
-					 errmsg("cannot create index on partitioned table \"%s\" concurrently",
-							RelationGetRelationName(rel))));
-	}
 
 	/*
 	 * Don't try to CREATE INDEX on temp tables of other backends.
@@ -1166,10 +1154,6 @@ DefineIndex(Oid tableId,
 		}
 	}
 
-	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
-	safe_index = indexInfo->ii_Expressions == NIL &&
-		indexInfo->ii_Predicate == NIL;
-
 	/*
 	 * Report index creation if appropriate (delay this till after most of the
 	 * error checks)
@@ -1234,6 +1218,11 @@ DefineIndex(Oid tableId,
 		if (pd->nparts != 0)
 			flags |= INDEX_CREATE_INVALID;
 	}
+	else if (concurrent && OidIsValid(parentIndexId))
+	{
+		/* If concurrent, initially build index partitions as "invalid" */
+		flags |= INDEX_CREATE_INVALID;
+	}
 
 	if (stmt->deferrable)
 		constr_flags |= INDEX_CONSTR_CREATE_DEFERRABLE;
@@ -1551,21 +1540,7 @@ DefineIndex(Oid tableId,
 			 */
 			if (invalidate_parent)
 			{
-				Relation	pg_index = table_open(IndexRelationId, RowExclusiveLock);
-				HeapTuple	tup,
-							newtup;
-
-				tup = SearchSysCache1(INDEXRELID,
-									  ObjectIdGetDatum(indexRelationId));
-				if (!HeapTupleIsValid(tup))
-					elog(ERROR, "cache lookup failed for index %u",
-						 indexRelationId);
-				newtup = heap_copytuple(tup);
-				((Form_pg_index) GETSTRUCT(newtup))->indisvalid = false;
-				CatalogTupleUpdate(pg_index, &tup->t_self, newtup);
-				ReleaseSysCache(tup);
-				table_close(pg_index, RowExclusiveLock);
-				heap_freetuple(newtup);
+				index_set_state_flags(indexRelationId, INDEX_DROP_CLEAR_VALID);
 
 				/*
 				 * CCI here to make this update visible, in case this recurses
@@ -1577,37 +1552,49 @@ DefineIndex(Oid tableId,
 
 		/*
 		 * Indexes on partitioned tables are not themselves built, so we're
-		 * done here.
+		 * done here in the non-concurrent case.
 		 */
-		AtEOXact_GUC(false, root_save_nestlevel);
-		SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
-		table_close(rel, NoLock);
-		if (!OidIsValid(parentIndexId))
-			pgstat_progress_end_command();
-		else
+		if (!concurrent)
 		{
-			/* Update progress for an intermediate partitioned index itself */
-			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
-		}
+			AtEOXact_GUC(false, root_save_nestlevel);
+			SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
+			table_close(rel, NoLock);
 
-		return address;
+			if (!OidIsValid(parentIndexId))
+				pgstat_progress_end_command();
+			else
+			{
+				/*
+				 * Update progress for an intermediate partitioned index
+				 * itself
+				 */
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+			}
+
+			return address;
+		}
 	}
 
 	AtEOXact_GUC(false, root_save_nestlevel);
 	SetUserIdAndSecContext(root_save_userid, root_save_sec_context);
 
-	if (!concurrent)
+	/*
+	 * All done in the non-concurrent case, and when building catalog entries
+	 * of partitions for CIC.
+	 */
+	if (!concurrent || OidIsValid(parentIndexId))
 	{
-		/* Close the heap and we're done, in the non-concurrent case */
 		table_close(rel, NoLock);
 
 		/*
 		 * If this is the top-level index, the command is done overall;
-		 * otherwise, increment progress to report one child index is done.
+		 * otherwise (when being called recursively), increment progress to
+		 * report that one child index is done.  Except in the concurrent
+		 * (catalog-only) case, which is handled later.
 		 */
 		if (!OidIsValid(parentIndexId))
 			pgstat_progress_end_command();
-		else
+		else if (!concurrent)
 			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
 
 		return address;
@@ -1618,6 +1605,144 @@ DefineIndex(Oid tableId,
 	SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
 	table_close(rel, NoLock);
 
+	if (!partitioned)
+	{
+		/* CREATE INDEX CONCURRENTLY on a nonpartitioned table */
+		DefineIndexConcurrentInternal(tableId, indexRelationId,
+									  indexInfo, heaplocktag, heaprelid);
+		pgstat_progress_end_command();
+		return address;
+	}
+	else
+	{
+		/*
+		 * For CIC on a partitioned table, finish by building indexes on
+		 * partitions
+		 */
+
+		ListCell   *lc;
+		List	   *childs;
+		List	   *tosetvalid = NIL;
+		MemoryContext cic_context,
+					old_context;
+		bool	pushed_snapshot = false;
+
+		/* Create special memory context for cross-transaction storage */
+		cic_context = AllocSetContextCreate(PortalContext,
+											"Create index concurrently",
+											ALLOCSET_DEFAULT_SIZES);
+
+		old_context = MemoryContextSwitchTo(cic_context);
+		childs = find_all_inheritors(indexRelationId, ShareUpdateExclusiveLock, NULL);
+		MemoryContextSwitchTo(old_context);
+
+		foreach(lc, childs)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Oid			tabrelid;
+			char		relkind;
+
+			/*
+			 * Partition could have been dropped, since we looked it up. In
+			 * this case consider it done and go to the next one.
+			 */
+			tabrelid = IndexGetRelation(indrelid, true);
+			if (!tabrelid)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+			rel = try_table_open(tabrelid, ShareUpdateExclusiveLock);
+			if (!rel)
+			{
+				pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+				continue;
+			}
+
+			/*
+			 * Pre-existing partitions which were ATTACHED were already
+			 * counted in the progress report.
+			 */
+			if (get_index_isvalid(indrelid))
+			{
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			/*
+			 * Partitioned indexes are counted in the progress report, but
+			 * don't need to be further processed.
+			 */
+			relkind = get_rel_relkind(indrelid);
+			if (!RELKIND_HAS_STORAGE(relkind))
+			{
+				/* The toplevel index doesn't count towards "partitions done" */
+				if (indrelid != indexRelationId)
+					pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+
+				/*
+				 * Build up a list of all the intermediate partitioned tables
+				 * which will later need to be set valid.
+				 */
+				old_context = MemoryContextSwitchTo(cic_context);
+				tosetvalid = lappend_oid(tosetvalid, indrelid);
+				MemoryContextSwitchTo(old_context);
+				table_close(rel, ShareUpdateExclusiveLock);
+				continue;
+			}
+
+			heaprelid = rel->rd_lockInfo.lockRelId;
+
+			/*
+			 * Close the table but retain the lock, that should be extended to
+			 * session level in DefineIndexConcurrentInternal.
+			 */
+			table_close(rel, NoLock);
+			SET_LOCKTAG_RELATION(heaplocktag, heaprelid.dbId, heaprelid.relId);
+
+			/* Process each partition in a separate transaction */
+			DefineIndexConcurrentInternal(tabrelid, indrelid, indexInfo,
+										  heaplocktag, heaprelid);
+
+			PushActiveSnapshot(GetTransactionSnapshot());
+			pushed_snapshot = true;
+			pgstat_progress_incr_param(PROGRESS_CREATEIDX_PARTITIONS_DONE, 1);
+		}
+
+		/* Set as valid all partitioned indexes, including the parent */
+		foreach(lc, tosetvalid)
+		{
+			Oid			indrelid = lfirst_oid(lc);
+			Relation	indrel = try_index_open(indrelid, ShareUpdateExclusiveLock);
+
+			if (!indrel)
+				continue;
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_READY);
+			CommandCounterIncrement();
+			index_set_state_flags(indrelid, INDEX_CREATE_SET_VALID);
+			index_close(indrel, ShareUpdateExclusiveLock);
+		}
+
+		MemoryContextDelete(cic_context);
+		pgstat_progress_end_command();
+		if (pushed_snapshot)
+			PopActiveSnapshot();
+		return address;
+	}
+}
+
+
+static void
+DefineIndexConcurrentInternal(Oid tableId, Oid indexRelationId, IndexInfo *indexInfo,
+							  LOCKTAG heaplocktag, LockRelId heaprelid)
+{
+	TransactionId limitXmin;
+	Snapshot	snapshot;
+
+	/* Is index safe for others to ignore?  See set_indexsafe_procflags() */
+	bool		safe_index = indexInfo->ii_Expressions == NIL &&
+		indexInfo->ii_Predicate == NIL;
+
 	/*
 	 * For a concurrent build, it's important to make the catalog entries
 	 * visible to other transactions before we start to build the index. That
@@ -1821,10 +1946,6 @@ DefineIndex(Oid tableId,
 	 * Last thing to do is release the session-level lock on the parent table.
 	 */
 	UnlockRelationIdForSession(&heaprelid, ShareUpdateExclusiveLock);
-
-	pgstat_progress_end_command();
-
-	return address;
 }
 
 
diff --git a/src/test/isolation/expected/partitioned-cic.out b/src/test/isolation/expected/partitioned-cic.out
new file mode 100644
index 00000000000..b66acc6f6a2
--- /dev/null
+++ b/src/test/isolation/expected/partitioned-cic.out
@@ -0,0 +1,135 @@
+Parsed test spec with 3 sessions
+
+starting permutation: lock_p1 cic insert drop2 commit chk_content
+step lock_p1: lock cictab_part_1 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step drop2: DROP TABLE cictab_part_2;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
+
+starting permutation: lock_p2 cic insert drop1 commit chk_content
+step lock_p2: lock cictab_part_2 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step drop1: DROP TABLE cictab_part_1;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+
+starting permutation: lock_p1 cic insert detach2 commit chk_content chk_content_part2
+step lock_p1: lock cictab_part_1 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step detach2: ALTER TABLE cictab DETACH PARTITION cictab_part_2;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
+step chk_content_part2: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_2 where i > 0;
+  select * from cictab_part_2 where i > 0;
+
+QUERY PLAN                                           
+-----------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2
+  Index Cond: (i > 0)                                
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+
+starting permutation: lock_p2 cic insert detach1 commit chk_content chk_content_part1
+step lock_p2: lock cictab_part_2 in row exclusive mode;
+step cic: CREATE INDEX CONCURRENTLY ON cictab(i); <waiting ...>
+step insert: insert into cictab values (1, 1), (11, 1);
+step detach1: ALTER TABLE cictab DETACH PARTITION cictab_part_1;
+step commit: COMMIT;
+step cic: <... completed>
+step chk_content: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+
+QUERY PLAN                                                  
+------------------------------------------------------------
+Index Scan using cictab_part_2_i_idx on cictab_part_2 cictab
+  Index Cond: (i > 0)                                       
+(2 rows)
+
+ i|j
+--+-
+11|0
+11|1
+(2 rows)
+
+step chk_content_part1: 
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_1 where i > 0;
+  select * from cictab_part_1 where i > 0;
+
+QUERY PLAN                                           
+-----------------------------------------------------
+Index Scan using cictab_part_1_i_idx on cictab_part_1
+  Index Cond: (i > 0)                                
+(2 rows)
+
+i|j
+-+-
+1|0
+1|1
+(2 rows)
+
diff --git a/src/test/isolation/isolation_schedule b/src/test/isolation/isolation_schedule
index e3c669a29c7..270efefdd71 100644
--- a/src/test/isolation/isolation_schedule
+++ b/src/test/isolation/isolation_schedule
@@ -116,3 +116,4 @@ test: serializable-parallel-2
 test: serializable-parallel-3
 test: matview-write-skew
 test: lock-nowait
+test: partitioned-cic
diff --git a/src/test/isolation/specs/partitioned-cic.spec b/src/test/isolation/specs/partitioned-cic.spec
new file mode 100644
index 00000000000..95f0bb2b47a
--- /dev/null
+++ b/src/test/isolation/specs/partitioned-cic.spec
@@ -0,0 +1,57 @@
+# Test the ability to drop/detach partitions while CREATE INDEX CONCURRENTLY is running.
+# To achieve this, start a transaction that will pause CIC in progress by
+# locking a partition in row exclusive mode, giving us a change to drop/detach another partition.
+# Dropping/detaching is tested for each partition to test two scenarios:
+# when the partition has already been indexed and when it's yet to be indexed.
+
+setup {
+  create table cictab(i int, j int) partition by range(i);
+  create table cictab_part_1 partition of cictab for values from (0) to (10);
+  create table cictab_part_2 partition of cictab for values from (10) to (20);
+
+  insert into cictab values (1, 0), (11, 0);
+}
+
+teardown {
+    drop table if exists cictab_part_1;
+    drop table if exists cictab_part_2;
+    drop table cictab;
+}
+
+session s1
+setup {BEGIN;}
+step lock_p1 { lock cictab_part_1 in row exclusive mode; }
+step lock_p2 { lock cictab_part_2 in row exclusive mode; }
+step commit { COMMIT; }
+
+session s2
+step cic { CREATE INDEX CONCURRENTLY ON cictab(i); }
+step chk_content {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab where i > 0;
+  select * from cictab where i > 0;
+}
+
+step chk_content_part1 {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_1 where i > 0;
+  select * from cictab_part_1 where i > 0;
+}
+
+step chk_content_part2 {
+  set enable_seqscan to off;
+  explain (costs off) select * from cictab_part_2 where i > 0;
+  select * from cictab_part_2 where i > 0;
+}
+
+session s3
+step detach1 { ALTER TABLE cictab DETACH PARTITION cictab_part_1; }
+step detach2 { ALTER TABLE cictab DETACH PARTITION cictab_part_2; }
+step drop1 { DROP TABLE cictab_part_1; }
+step drop2 { DROP TABLE cictab_part_2; }
+step insert { insert into cictab values (1, 1), (11, 1); }
+
+permutation lock_p1 cic insert drop2 commit chk_content
+permutation lock_p2 cic insert drop1 commit chk_content
+permutation lock_p1 cic insert detach2 commit chk_content chk_content_part2
+permutation lock_p2 cic insert detach1 commit chk_content chk_content_part1
diff --git a/src/test/regress/expected/indexing.out b/src/test/regress/expected/indexing.out
index bcf1db11d73..ff448e0f988 100644
--- a/src/test/regress/expected/indexing.out
+++ b/src/test/regress/expected/indexing.out
@@ -50,12 +50,136 @@ select relname, relkind, relhassubclass, inhparent::regclass
 (8 rows)
 
 drop table idxpart;
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-ERROR:  cannot create index on partitioned table "idxpart" concurrently
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+ERROR:  could not create unique index "idxpart2_a_idx1"
+DETAIL:  Key (a)=(10) is duplicated.
+\d idxpart
+        Partitioned table "public.idxpart"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition key: RANGE (a)
+Indexes:
+    "idxpart_a_idx" btree (a)
+Number of partitions: 3 (Use \d+ to list them.)
+
+\d idxpart1
+        Partitioned table "public.idxpart1"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1_a_idx" btree (a)
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart11
+       Partitioned table "public.idxpart11"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart1 FOR VALUES FROM (0) TO (10)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart11_a_idx" btree (a)
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart111
+       Partitioned table "public.idxpart111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart11 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart111_a_idx" btree (a)
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart1111
+      Partitioned table "public.idxpart1111"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart111 DEFAULT
+Partition key: RANGE (a)
+Indexes:
+    "idxpart1111_a_idx" btree (a)
+Number of partitions: 0
+
+\d idxpart2
+              Table "public.idxpart2"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (10) TO (20)
+Indexes:
+    "idxpart2_a_idx" btree (a)
+
+\d idxpart3
+        Partitioned table "public.idxpart3"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart FOR VALUES FROM (30) TO (40)
+Partition key: RANGE (a)
+Indexes:
+    "idxpart3_a_idx" btree (a)
+Number of partitions: 1 (Use \d+ to list them.)
+
+\d idxpart31
+             Table "public.idxpart31"
+ Column |  Type   | Collation | Nullable | Default 
+--------+---------+-----------+----------+---------
+ a      | integer |           |          | 
+ b      | integer |           |          | 
+ c      | text    |           |          | 
+Partition of: idxpart3 DEFAULT
+Indexes:
+    "idxpart31_a_idx" btree (a)
+
+drop table idxpart;
+-- CIC should preserve active snapshot
+create function access_snapshot() returns event_trigger language plpgsql as $$
+begin
+perform 1;
+end
+$$;
+create table idxpart (col1 int) partition by range (col1);
+create event trigger check_trigger on ddl_command_end when tag in ('CREATE INDEX')
+execute procedure access_snapshot();
+create index concurrently on idxpart (col1);
+drop event trigger check_trigger;
 drop table idxpart;
+drop function access_snapshot();
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
 CREATE TABLE idxpart (col1 INT) PARTITION BY RANGE (col1);
diff --git a/src/test/regress/sql/indexing.sql b/src/test/regress/sql/indexing.sql
index b5cb01c2d70..b6b76424e76 100644
--- a/src/test/regress/sql/indexing.sql
+++ b/src/test/regress/sql/indexing.sql
@@ -29,11 +29,45 @@ select relname, relkind, relhassubclass, inhparent::regclass
 	where relname like 'idxpart%' order by relname;
 drop table idxpart;
 
--- Some unsupported features
+-- CIC on partitioned table
 create table idxpart (a int, b int, c text) partition by range (a);
-create table idxpart1 partition of idxpart for values from (0) to (10);
-create index concurrently on idxpart (a);
-drop table idxpart;
+create table idxpart1 partition of idxpart for values from (0) to (10) partition by range(a);
+create table idxpart11 partition of idxpart1 for values from (0) to (10) partition by range(a);
+create table idxpart111 partition of idxpart11 default partition by range(a);
+create table idxpart1111 partition of idxpart111 default partition by range(a);
+create table idxpart2 partition of idxpart for values from (10) to (20);
+create table idxpart3 partition of idxpart for values from (30) to (40) partition by range(a);
+create table idxpart31 partition of idxpart3 default;
+
+insert into idxpart2 values(10),(10); -- not unique
+create index concurrently on idxpart11 (a); -- partitioned and partition, with no leaves
+create index concurrently on idxpart1 (a); -- partitioned and partition
+create index concurrently on idxpart2 (a); -- leaf
+create index concurrently on idxpart (a); -- partitioned
+create unique index concurrently on idxpart (a); -- partitioned, unique failure
+\d idxpart
+\d idxpart1
+\d idxpart11
+\d idxpart111
+\d idxpart1111
+\d idxpart2
+\d idxpart3
+\d idxpart31
+drop table idxpart;
+-- CIC should preserve active snapshot
+create function access_snapshot() returns event_trigger language plpgsql as $$
+begin
+perform 1;
+end
+$$;
+create table idxpart (col1 int) partition by range (col1);
+create event trigger check_trigger on ddl_command_end when tag in ('CREATE INDEX')
+execute procedure access_snapshot();
+create index concurrently on idxpart (col1);
+drop event trigger check_trigger;
+drop table idxpart;
+drop function access_snapshot();
+
 
 -- Verify bugfix with query on indexed partitioned table with no partitions
 -- https://postgr.es/m/20180124162006.pmapfiznhgngwtjf@alvherre.pgsql
-- 
2.43.0