[Proposal] Add accumulated statistics for wait event

From 758cdc0d52550e6930470a8cf6936269b714539c Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Wed, 30 Oct 2019 05:36:50 +0000
Subject: [PATCH 1/2] Adding the pg_stat_waitaccum view which shows counts and
 duration of each wait events. Each backend/backgrounds counts and measures
 the time of wait event in every pgstat_report_wait_start and
 pgstat_report_wait_end. They store those info into their local variables and
 send to Statistics Collector. We can get those info via Statistics Collector.

For reducing overhead, I implemented statistic hash instead of
dynamic hash. I also implemented track_wait_timing which
determines wait event duration is collected or not.

On windows, this function might be not worked correctly, because
now it initialize local variables in pg_stat_init which is not
passed to fork processes on windows.
---
 src/backend/catalog/system_views.sql          |   8 +
 src/backend/postmaster/pgstat.c               | 344 ++++++++++++++++++++++++++
 src/backend/storage/lmgr/lwlock.c             |  19 ++
 src/backend/utils/adt/pgstatfuncs.c           |  79 ++++++
 src/backend/utils/misc/guc.c                  |   9 +
 src/backend/utils/misc/postgresql.conf.sample |   1 +
 src/include/catalog/pg_proc.dat               |   9 +
 src/include/pgstat.h                          | 123 ++++++++-
 src/include/storage/lwlock.h                  |   1 +
 src/include/storage/proc.h                    |   1 +
 src/test/regress/expected/rules.out           |   5 +
 11 files changed, 597 insertions(+), 2 deletions(-)

diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index 9fe4a47..1ca0bbe 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -946,6 +946,14 @@ CREATE VIEW pg_stat_bgwriter AS
         pg_stat_get_buf_alloc() AS buffers_alloc,
         pg_stat_get_bgwriter_stat_reset_time() AS stats_reset;
 
+CREATE VIEW pg_stat_waitaccum AS
+    SELECT
+		S.wait_event_type AS wait_event_type,
+		S.wait_event AS wait_event,
+		S.calls AS calls,
+		S.times AS times
+	FROM pg_stat_get_waitaccum(NULL) AS S;
+
 CREATE VIEW pg_stat_progress_vacuum AS
     SELECT
         S.pid AS pid, S.datid AS datid, D.datname AS datname,
diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 011076c..5a07a81 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -124,6 +124,7 @@
  */
 bool		pgstat_track_activities = false;
 bool		pgstat_track_counts = false;
+bool		pgstat_track_wait_timing = false;
 int			pgstat_track_functions = TRACK_FUNC_OFF;
 int			pgstat_track_activity_query_size = 1024;
 
@@ -154,6 +155,10 @@ static time_t last_pgstat_start_time;
 
 static bool pgStatRunningInCollector = false;
 
+WAHash *wa_hash;
+
+instr_time waitStart;
+
 /*
  * Structures in which backends store per-table info that's waiting to be
  * sent to the collector.
@@ -256,6 +261,7 @@ static int	localNumBackends = 0;
  */
 static PgStat_ArchiverStats archiverStats;
 static PgStat_GlobalStats globalStats;
+static PgStat_WaitAccumStats waitAccumStats;
 
 /*
  * List of OIDs of databases we need to write out.  If an entry is InvalidOid,
@@ -285,6 +291,8 @@ static pid_t pgstat_forkexec(void);
 #endif
 
 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
+static void pgstat_init_waitaccum_hash(WAHash **hash);
+static PgStat_WaitAccumEntry *pgstat_add_wa_entry(WAHash *hash, uint32 key);
 static void pgstat_exit(SIGNAL_ARGS);
 static void pgstat_beshutdown_hook(int code, Datum arg);
 static void pgstat_sighup_handler(SIGNAL_ARGS);
@@ -294,8 +302,11 @@ static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
 												 Oid tableoid, bool create);
 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
+static void pgstat_write_waitaccum_statsfile(FILE *fpout);
 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
+static bool pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+											FILE *fpin, const char *statfile);
 static void backend_read_statsfile(void);
 static void pgstat_read_current_status(void);
 
@@ -331,6 +342,7 @@ static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
+static void pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len);
 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
@@ -338,6 +350,27 @@ static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
 static void pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len);
 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
 
+
+PgStat_WaitAccumEntry *
+pgstat_get_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *current;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	current = hash->buckets[bucket];
+
+	while (current != NULL)
+	{
+		if (current->key == key)
+			return current->entry;
+
+		current = current->next;
+	}
+
+	return NULL;
+}
+
+
 /* ------------------------------------------------------------
  * Public functions called from postmaster follow
  * ------------------------------------------------------------
@@ -609,6 +642,8 @@ retry2:
 
 	pg_freeaddrinfo_all(hints.ai_family, addrs);
 
+	pgstat_init_waitaccum_hash(&wa_hash);
+
 	return;
 
 startup_failed:
@@ -631,6 +666,75 @@ startup_failed:
 	SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
 }
 
+static PgStat_WaitAccumEntry *
+pgstat_add_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *prev;
+	WAEntry *new;
+	int bucket = key % WA_BUCKET_SIZE;
+	
+	prev = hash->buckets[bucket];
+
+	while (prev != NULL && prev->next != NULL)
+		prev = prev->next;
+	
+	new = &hash->entries[hash->entry_num++];
+	new->key = key;
+	new->entry = MemoryContextAllocZero(TopMemoryContext, (sizeof(PgStat_WaitAccumEntry)));
+
+	if (prev != NULL)
+		prev->next = new;
+	else
+		hash->buckets[bucket] = new;
+
+	return new->entry;
+}
+
+static void
+pgstat_init_waitaccum_entry(WAHash *hash, uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+
+	entry = pgstat_add_wa_entry(hash, wait_event_info);
+	entry->wait_event_info = wait_event_info;
+}
+
+static void
+pgstat_init_waitaccum_hash(WAHash **hash)
+{
+	uint32 i;
+	int last_tranche_id;
+
+	*hash = MemoryContextAllocZero(TopMemoryContext, sizeof(WAHash));
+
+	last_tranche_id = LWLockGetLastTrancheId();
+	for (i = PG_WAIT_LWLOCK + 1; i <= (PG_WAIT_LWLOCK | last_tranche_id); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = (PG_WAIT_LOCK | LOCKTAG_RELATION); i <= (PG_WAIT_LOCK | LOCKTAG_LAST_TYPE); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_BUFFER_PIN; i <= PG_WAIT_BUFFER_PIN; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_ACTIVITY; i <= PG_WAIT_ACTIVITY_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_CLIENT; i <= PG_WAIT_CLIENT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	//do extension stuff
+
+	for (i = PG_WAIT_IPC; i <= PG_WAIT_IPC_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_TIMEOUT; i <= PG_WAIT_TIMEOUT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IO; i <= PG_WAIT_IO_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+}
+
 /*
  * subroutine for pgstat_reset_all
  */
@@ -911,6 +1015,9 @@ pgstat_report_stat(bool force)
 
 	/* Now, send function statistics */
 	pgstat_send_funcstats();
+
+	/* Send wait accumulative statistics */
+	pgstat_send_waitaccum();
 }
 
 /*
@@ -1341,6 +1448,8 @@ pgstat_reset_shared_counters(const char *target)
 		msg.m_resettarget = RESET_ARCHIVER;
 	else if (strcmp(target, "bgwriter") == 0)
 		msg.m_resettarget = RESET_BGWRITER;
+	else if (strcmp(target, "waitaccum") == 0)
+		msg.m_resettarget = RESET_WAITACCUM;
 	else
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
@@ -2625,6 +2734,22 @@ pgstat_fetch_global(void)
 	return &globalStats;
 }
 
+/*
+ * ---------
+ * pgstat_fetch_stat_waitaccum() -
+ *
+ *	Support function for the SQL-callable pgstat* functions. Returns
+ *	a pointer to the wait accum statistics struct.
+ * ---------
+ */
+PgStat_WaitAccumStats *
+pgstat_fetch_stat_waitaccum(void)
+{
+	backend_read_statsfile();
+
+	return &waitAccumStats;
+}
+
 
 /* ------------------------------------------------------------
  * Functions for management of the shared-memory PgBackendStatus array
@@ -4414,6 +4539,53 @@ pgstat_send_bgwriter(void)
 	MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
 }
 
+/* ----------
+ * pgstat_send_waitaccum() -
+ *
+ * ----------
+ */
+void
+pgstat_send_waitaccum()
+{
+	PgStat_MsgWaitAccum msg;
+	PgStat_WaitAccumEntry *entry;
+	int i;
+
+	if (wa_hash == NULL)
+		return;
+
+	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_WAITACCUM);
+	msg.m_nentries = 0;
+
+	for (i = 0; i < wa_hash->entry_num; i++)
+	{
+		entry = wa_hash->entries[i].entry;
+
+		/* Send only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Prepare and send the message
+		 */
+		memcpy(&msg.m_entry[msg.m_nentries], entry, sizeof(PgStat_WaitAccumEntry));
+		if (++msg.m_nentries >= PGSTAT_NUM_WAITACCUMENTRIES)
+		{
+			pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+						msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+			msg.m_nentries = 0;
+		}
+
+		/* Clear wait events information. */
+		entry->calls = 0;
+		INSTR_TIME_SET_ZERO(entry->times);
+	}
+
+	if (msg.m_nentries > 0)
+		pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+					msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+}
+
 
 /* ----------
  * PgstatCollectorMain() -
@@ -4606,6 +4778,10 @@ PgstatCollectorMain(int argc, char *argv[])
 					pgstat_recv_bgwriter(&msg.msg_bgwriter, len);
 					break;
 
+				case PGSTAT_MTYPE_WAITACCUM:
+					pgstat_recv_waitaccum(&msg.msg_waitaccum, len);
+					break;
+
 				case PGSTAT_MTYPE_FUNCSTAT:
 					pgstat_recv_funcstat(&msg.msg_funcstat, len);
 					break;
@@ -4901,6 +5077,8 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 	rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
 	(void) rc;					/* we'll check for error with ferror */
 
+	pgstat_write_waitaccum_statsfile(fpout);
+
 	/*
 	 * Walk through the database table.
 	 */
@@ -5106,6 +5284,43 @@ pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
 }
 
 /* ----------
+ * pgstat_write_waitaccum_statsfile() -
+ *		Write the waitAccumStats to the stat file.
+ *
+ * ----------
+ */
+static void
+pgstat_write_waitaccum_statsfile(FILE *fpout)
+{
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			rc;
+	int			i;
+
+	/*
+	 * Walk through the waitaccum hash.
+	 */
+	for (i = 0; i < hash->entry_num; i++)
+	{
+		entry = hash->entries[i].entry;
+
+		/* Write only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Write out the DB entry. We don't write the tables or functions
+		 * pointers, since they're of no use to any other process.
+		 */
+		fputc('D', fpout);
+		rc = fwrite(entry, sizeof(PgStat_WaitAccumEntry), 1, fpout);
+		(void) rc;				/* we'll check for error with ferror */
+	}	
+
+	fputc('E', fpout);
+}
+
+/* ----------
  * pgstat_read_statsfiles() -
  *
  *	Reads in some existing statistics collector files and returns the
@@ -5158,6 +5373,7 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 	 */
 	memset(&globalStats, 0, sizeof(globalStats));
 	memset(&archiverStats, 0, sizeof(archiverStats));
+	waitAccumStats.hash = MemoryContextAllocZero(pgStatLocalContext, sizeof(WAHash));
 
 	/*
 	 * Set the current timestamp (will be kept only in case we can't load an
@@ -5228,6 +5444,9 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 		goto done;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&waitAccumStats, fpin, statfile))
+		goto done;
+
 	/*
 	 * We found an existing collector stats file. Read it and put all the
 	 * hashtable entries into place.
@@ -5526,10 +5745,13 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 	PgStat_StatDBEntry dbentry;
 	PgStat_GlobalStats myGlobalStats;
 	PgStat_ArchiverStats myArchiverStats;
+	PgStat_WaitAccumStats myWaitAccumStats;
 	FILE	   *fpin;
 	int32		format_id;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
 
+	myWaitAccumStats.hash = MemoryContextAllocZero(CurrentMemoryContext, sizeof(WAHash));
+
 	/*
 	 * Try to open the stats file.  As above, anything but ENOENT is worthy of
 	 * complaining about.
@@ -5580,6 +5802,9 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 		return false;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&myWaitAccumStats, fpin, statfile))
+		return false;
+
 	/* By default, we're going to return the timestamp of the global file. */
 	*ts = myGlobalStats.stats_timestamp;
 
@@ -5633,6 +5858,75 @@ done:
 	return true;
 }
 
+/* ----------
+ * pgstat_read_statsfiles() -
+ *
+ *	Reads the waitaccum stats from the file.
+ *	If an error happens when reading file, return false. Otherwise return true.
+ *
+ * ----------
+ */
+static bool
+pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+								FILE *fpin, const char *statfile)
+{
+	PgStat_WaitAccumEntry *entry;
+	PgStat_WaitAccumEntry buf;
+	WAHash *hash = stats->hash;
+
+	/*
+	 * Read and put all the hashtable entries into place.
+	 */
+	for (;;)
+	{
+		switch (fgetc(fpin))
+		{
+				/*
+				 * 'D'	A PgStat_WaitAccumEntry struct describing a database
+				 * follows.
+				 */
+			case 'D':
+				if (fread(&buf, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+									 != sizeof(PgStat_WaitAccumEntry))
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				entry = pgstat_get_wa_entry(hash, buf.wait_event_info);
+
+				if (entry)
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				/*
+				 * Add to the DB hash
+				 */
+				entry = pgstat_add_wa_entry(hash, buf.wait_event_info);
+				memcpy(entry, &buf, sizeof(PgStat_WaitAccumEntry));
+
+				break;
+
+			case 'E':
+				return true;
+
+			default:
+				ereport(pgStatRunningInCollector ? LOG : WARNING,
+						(errmsg("corrupted statistics file \"%s\"",
+								statfile)));
+				return false;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * If not already done, read the statistics collector stats file into
  * some hash tables.  The results will be kept until pgstat_clear_snapshot()
@@ -6142,7 +6436,20 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 		memset(&archiverStats, 0, sizeof(archiverStats));
 		archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
 	}
+	else if (msg->m_resettarget == RESET_WAITACCUM)
+	{
+		PgStat_WaitAccumEntry *entry;
+		WAHash *hash = waitAccumStats.hash;
+		int i;
+
+		for (i = 0; i < hash->entry_num; i++)
+		{
+			entry = hash->entries[i].entry;
 
+			entry->calls = 0;
+			INSTR_TIME_SET_ZERO(entry->times);
+		}
+	}
 	/*
 	 * Presumably the sender of this message validated the target, don't
 	 * complain here if it's not valid
@@ -6322,6 +6629,43 @@ pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
 }
 
 /* ----------
+ * pgstat_recv_waitaccum() -
+ *
+ *	Process a WAITACCUM message.
+ * ----------
+ */
+static void
+pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
+{
+	PgStat_WaitAccumEntry *m_entry = &(msg->m_entry[0]);
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			i;
+
+	/*
+	 * Process all function entries in the message.
+	 */
+	for (i = 0; i < msg->m_nentries; i++, m_entry++)
+	{
+		entry = pgstat_get_wa_entry(hash, m_entry->wait_event_info);
+
+		if (!entry)
+		{
+			entry = pgstat_add_wa_entry(hash, m_entry->wait_event_info);
+			memcpy(entry, m_entry, sizeof(PgStat_WaitAccumEntry));
+		}
+		else
+		{
+			/*
+			 * Otherwise add the values to the existing entry.
+			 */
+			entry->calls += m_entry->calls;
+			INSTR_TIME_ADD(entry->times, m_entry->times);
+		}
+	}
+}
+
+/* ----------
  * pgstat_recv_recoveryconflict() -
  *
  *	Process a RECOVERYCONFLICT message.
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index fb0bf44..397d455 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -594,6 +594,25 @@ LWLockNewTrancheId(void)
 }
 
 /*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
+/*
  * Register a tranche ID in the lookup table for the current process.  This
  * routine will save a pointer to the tranche name passed as an argument,
  * so the name should be allocated in a backend-lifetime context
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 05240bf..8864fe1 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -1970,3 +1970,82 @@ pg_stat_get_archiver(PG_FUNCTION_ARGS)
 	PG_RETURN_DATUM(HeapTupleGetDatum(
 									  heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	PgStat_WaitAccumStats *waitaccum_stats;
+	PgStat_WaitAccumEntry *entry;
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	/* Get statistics about the waitaccum process */
+	waitaccum_stats = pgstat_fetch_stat_waitaccum();
+
+	for (i = 0; i < waitaccum_stats->hash->entry_num; i++)
+	{
+		entry = waitaccum_stats->hash->entries[i].entry;
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 31a5ef0..1fb088e 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1407,6 +1407,15 @@ static struct config_bool ConfigureNamesBool[] =
 		false,
 		NULL, NULL, NULL
 	},
+	{
+		{"track_wait_timing", PGC_SUSET, STATS_COLLECTOR,
+			gettext_noop("Collects timing statistics for wait events."),
+			NULL
+		},
+		&pgstat_track_wait_timing,
+		false,
+		NULL, NULL, NULL
+	},
 
 	{
 		{"update_process_title", PGC_SUSET, PROCESS_TITLE,
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index 0fc23e3..52f511c 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -556,6 +556,7 @@
 #track_activities = on
 #track_counts = on
 #track_io_timing = off
+#track_wait_timing = off
 #track_functions = none			# none, pl, all
 #track_activity_query_size = 1024	# (change requires restart)
 #stats_temp_directory = 'pg_stat_tmp'
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index 58ea5b9..487bc65 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5152,6 +5152,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,sslcompression,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '2228',
+  descr => 'statistics: information about accumulative data of wait event',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 's', proparallel => 'r',
+  prorettype => 'record', proargtypes => 'int4',
+  proallargtypes => '{int4,text,text,int8,int8}',
+  proargmodes => '{i,o,o,o,o}',
+  proargnames => '{pid,wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index fe076d8..32c4b5f 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -59,6 +59,7 @@ typedef enum StatMsgType
 	PGSTAT_MTYPE_ANALYZE,
 	PGSTAT_MTYPE_ARCHIVER,
 	PGSTAT_MTYPE_BGWRITER,
+	PGSTAT_MTYPE_WAITACCUM,
 	PGSTAT_MTYPE_FUNCSTAT,
 	PGSTAT_MTYPE_FUNCPURGE,
 	PGSTAT_MTYPE_RECOVERYCONFLICT,
@@ -119,7 +120,8 @@ typedef struct PgStat_TableCounts
 typedef enum PgStat_Shared_Reset_Target
 {
 	RESET_ARCHIVER,
-	RESET_BGWRITER
+	RESET_BGWRITER,
+	RESET_WAITACCUM
 } PgStat_Shared_Reset_Target;
 
 /* Possible object types for resetting single counters */
@@ -423,6 +425,33 @@ typedef struct PgStat_MsgBgWriter
 } PgStat_MsgBgWriter;
 
 /* ----------
+ * PgStat_WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct PgStat_WaitAccumEntry
+{
+	uint32			wait_event_info;
+	PgStat_Counter	calls;
+	instr_time		times;
+} PgStat_WaitAccumEntry;
+
+/* ----------
+ * PgStat_MsgWaitAccum	Sent by backend/background's process to update statistics.
+ * ----------
+ */
+#define PGSTAT_NUM_WAITACCUMENTRIES	\
+	((PGSTAT_MSG_PAYLOAD - sizeof(int))  \
+	 / sizeof(PgStat_WaitAccumEntry))
+
+typedef struct PgStat_MsgWaitAccum
+{
+	PgStat_MsgHdr m_hdr;
+
+	int m_nentries;
+	PgStat_WaitAccumEntry m_entry[PGSTAT_NUM_WAITACCUMENTRIES];
+} PgStat_MsgWaitAccum;
+
+/* ----------
  * PgStat_MsgRecoveryConflict	Sent by the backend upon recovery conflict
  * ----------
  */
@@ -564,6 +593,7 @@ typedef union PgStat_Msg
 	PgStat_MsgAnalyze msg_analyze;
 	PgStat_MsgArchiver msg_archiver;
 	PgStat_MsgBgWriter msg_bgwriter;
+	PgStat_MsgWaitAccum msg_waitaccum;
 	PgStat_MsgFuncstat msg_funcstat;
 	PgStat_MsgFuncpurge msg_funcpurge;
 	PgStat_MsgRecoveryConflict msg_recoveryconflict;
@@ -581,7 +611,7 @@ typedef union PgStat_Msg
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9D
+#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9E
 
 /* ----------
  * PgStat_StatDBEntry			The collector's data per database
@@ -711,6 +741,30 @@ typedef struct PgStat_GlobalStats
 	TimestampTz stat_reset_timestamp;
 } PgStat_GlobalStats;
 
+typedef struct WAEntry
+{
+	int key;
+	PgStat_WaitAccumEntry *entry;
+	struct WAEntry *next;
+} WAEntry;
+
+#define WA_BUCKET_SIZE 461
+
+typedef struct WAHash
+{
+	WAEntry entries[WA_BUCKET_SIZE];
+	WAEntry *buckets[WA_BUCKET_SIZE];
+	int entry_num;
+} WAHash;
+
+/*
+ * WaitAccum statistics kept in the stats collector
+ */
+typedef struct PgStat_WaitAccumStats
+{
+	WAHash *hash;
+} PgStat_WaitAccumStats;
+
 
 /* ----------
  * Backend types
@@ -787,6 +841,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+
 /* ----------
  * Wait Events - Client
  *
@@ -808,6 +864,8 @@ typedef enum
 	WAIT_EVENT_GSS_OPEN_SERVER,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_GSS_OPEN_SERVER
+
 /* ----------
  * Wait Events - IPC
  *
@@ -856,6 +914,8 @@ typedef enum
 	WAIT_EVENT_SYNC_REP
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_SYNC_REP
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -869,6 +929,8 @@ typedef enum
 	WAIT_EVENT_RECOVERY_APPLY_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_RECOVERY_APPLY_DELAY
+
 /* ----------
  * Wait Events - IO
  *
@@ -947,6 +1009,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_WAL_WRITE
+
 /* ----------
  * Command type for progress reporting purposes
  * ----------
@@ -1203,6 +1267,8 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
+extern WAHash *wa_hash;
+extern instr_time waitStart;
 
 /* ----------
  * GUC parameters
@@ -1210,6 +1276,7 @@ typedef struct PgStat_FunctionCallUsage
  */
 extern bool pgstat_track_activities;
 extern bool pgstat_track_counts;
+extern bool pgstat_track_wait_timing;
 extern int	pgstat_track_functions;
 extern PGDLLIMPORT int pgstat_track_activity_query_size;
 extern char *pgstat_stat_directory;
@@ -1227,6 +1294,7 @@ extern PgStat_MsgBgWriter BgWriterStats;
 extern PgStat_Counter pgStatBlockReadTime;
 extern PgStat_Counter pgStatBlockWriteTime;
 
+extern PgStat_WaitAccumEntry *pgstat_get_wa_entry(WAHash *hash, uint32 key);
 /* ----------
  * Functions called from postmaster
  * ----------
@@ -1314,6 +1382,50 @@ extern char *pgstat_clip_activity(const char *raw_activity);
  * initialized.
  * ----------
  */
+
+static inline void
+pgstat_report_waitaccum_start()
+{
+	if (wa_hash == NULL)
+		return; 
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(waitStart);
+	}
+}
+
+static inline void
+pgstat_report_waitaccum_end(uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+	instr_time  diff;
+
+	if (wa_hash == NULL)
+		return; 
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(diff);
+		INSTR_TIME_SUBTRACT(diff, waitStart);
+	}
+
+	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
+
+	if (!entry)
+	{
+		printf("wait_event_info: %u.\n", wait_event_info);
+		fflush(stdout);
+		return;
+	}
+
+	entry->calls++;
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+}
+
 static inline void
 pgstat_report_wait_start(uint32 wait_event_info)
 {
@@ -1327,6 +1439,8 @@ pgstat_report_wait_start(uint32 wait_event_info)
 	 * four-bytes, updates are atomic.
 	 */
 	proc->wait_event_info = wait_event_info;
+
+	pgstat_report_waitaccum_start();
 }
 
 /* ----------
@@ -1346,6 +1460,8 @@ pgstat_report_wait_end(void)
 	if (!pgstat_track_activities || !proc)
 		return;
 
+	pgstat_report_waitaccum_end(proc->wait_event_info);
+
 	/*
 	 * Since this is a four-byte field which is always read and written as
 	 * four-bytes, updates are atomic.
@@ -1353,6 +1469,7 @@ pgstat_report_wait_end(void)
 	proc->wait_event_info = 0;
 }
 
+
 /* nontransactional event counts are simple enough to inline */
 
 #define pgstat_count_heap_scan(rel)									\
@@ -1420,6 +1537,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 
 extern void pgstat_send_archiver(const char *xlog, bool failed);
 extern void pgstat_send_bgwriter(void);
+extern void pgstat_send_waitaccum(void);
 
 /* ----------
  * Support functions for the SQL-callable functions to
@@ -1434,5 +1552,6 @@ extern PgStat_StatFuncEntry *pgstat_fetch_stat_funcentry(Oid funcid);
 extern int	pgstat_fetch_stat_numbackends(void);
 extern PgStat_ArchiverStats *pgstat_fetch_stat_archiver(void);
 extern PgStat_GlobalStats *pgstat_fetch_global(void);
+extern PgStat_WaitAccumStats *pgstat_fetch_stat_waitaccum(void);
 
 #endif							/* PGSTAT_H */
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index f627dfe..5c69e08 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -184,6 +184,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 281e1db..4d03d80 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -21,6 +21,7 @@
 #include "storage/lock.h"
 #include "storage/pg_sema.h"
 #include "storage/proclist_types.h"
+#include "portability/instr_time.h"
 
 /*
  * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 210e9cd..da9c288 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2053,6 +2053,11 @@ pg_stat_user_tables| SELECT pg_stat_all_tables.relid,
     pg_stat_all_tables.autoanalyze_count
    FROM pg_stat_all_tables
   WHERE ((pg_stat_all_tables.schemaname <> ALL (ARRAY['pg_catalog'::name, 'information_schema'::name])) AND (pg_stat_all_tables.schemaname !~ '^pg_toast'::text));
+pg_stat_waitaccum| SELECT s.wait_event_type,
+    s.wait_event,
+    s.calls,
+    s.times
+   FROM pg_stat_get_waitaccum(NULL::integer) s(wait_event_type, wait_event, calls, times);
 pg_stat_wal_receiver| SELECT s.pid,
     s.status,
     s.receive_start_lsn,
-- 
1.8.3.1

From 5c3421d9634101fda4646d27681e4c9d14bca176 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Wed, 30 Oct 2019 05:38:11 +0000
Subject: [PATCH 2/2] [POC] Changed measuring method of wait event timed from
 INSTR_TIME (which uses gettimeofday or clock_gettime) to rdtsc. This might
 reduce the overhead of measuring overhead.

Any supports like changing clock cycle to actual time or error
handling are not currently implemented.
---
 src/backend/postmaster/pgstat.c      |  8 ++++----
 src/backend/utils/adt/pgstatfuncs.c  |  2 +-
 src/include/pgstat.h                 | 14 +++++++-------
 src/include/portability/instr_time.h | 23 +++++++++++++++++++++++
 4 files changed, 35 insertions(+), 12 deletions(-)

diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 5a07a81..b677042 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -157,7 +157,7 @@ static bool pgStatRunningInCollector = false;
 
 WAHash *wa_hash;
 
-instr_time waitStart;
+uint64 waitStart;
 
 /*
  * Structures in which backends store per-table info that's waiting to be
@@ -4578,7 +4578,7 @@ pgstat_send_waitaccum()
 
 		/* Clear wait events information. */
 		entry->calls = 0;
-		INSTR_TIME_SET_ZERO(entry->times);
+		entry->times = 0;
 	}
 
 	if (msg.m_nentries > 0)
@@ -6447,7 +6447,7 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 			entry = hash->entries[i].entry;
 
 			entry->calls = 0;
-			INSTR_TIME_SET_ZERO(entry->times);
+			entry->times = 0;
 		}
 	}
 	/*
@@ -6660,7 +6660,7 @@ pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
 			 * Otherwise add the values to the existing entry.
 			 */
 			entry->calls += m_entry->calls;
-			INSTR_TIME_ADD(entry->times, m_entry->times);
+			entry->times += m_entry->times;
 		}
 	}
 }
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 8864fe1..5ea4140 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -2039,7 +2039,7 @@ pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
 
 		values[2] = Int64GetDatum(entry->calls);
 
-		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+		values[3] = UInt64GetDatum(entry->times);
 
 		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
 	}
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 32c4b5f..2515893 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -432,7 +432,7 @@ typedef struct PgStat_WaitAccumEntry
 {
 	uint32			wait_event_info;
 	PgStat_Counter	calls;
-	instr_time		times;
+	uint64			times;
 } PgStat_WaitAccumEntry;
 
 /* ----------
@@ -1268,7 +1268,7 @@ typedef struct PgStat_FunctionCallUsage
 } PgStat_FunctionCallUsage;
 
 extern WAHash *wa_hash;
-extern instr_time waitStart;
+extern uint64 waitStart;
 
 /* ----------
  * GUC parameters
@@ -1391,7 +1391,7 @@ pgstat_report_waitaccum_start()
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(waitStart);
+		waitStart = rdtsc();
 	}
 }
 
@@ -1399,15 +1399,15 @@ static inline void
 pgstat_report_waitaccum_end(uint32 wait_event_info)
 {
 	PgStat_WaitAccumEntry *entry;
-	instr_time  diff;
+	uint64		diff = 0;
 
 	if (wa_hash == NULL)
 		return; 
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(diff);
-		INSTR_TIME_SUBTRACT(diff, waitStart);
+		diff = rdtsc();
+		diff -= waitStart;
 	}
 
 	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
@@ -1422,7 +1422,7 @@ pgstat_report_waitaccum_end(uint32 wait_event_info)
 	entry->calls++;
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_ADD(entry->times, diff);
+		entry->times += diff;
 	}
 }
 
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 0f5c161..668fa63 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -57,6 +57,10 @@
 
 #ifndef WIN32
 
+#if defined(__x86_64__) || defined(__i386__)
+#include <x86intrin.h>
+#endif
+
 #ifdef HAVE_CLOCK_GETTIME
 
 /* Use clock_gettime() */
@@ -209,6 +213,8 @@ typedef struct timeval instr_time;
 
 #else							/* WIN32 */
 
+#include <intrin.h>
+
 /* Use QueryPerformanceCounter() */
 
 typedef LARGE_INTEGER instr_time;
@@ -254,3 +260,20 @@ GetTimerFrequency(void)
 	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
 
 #endif							/* INSTR_TIME_H */
+
+
+#ifndef RDTSC_H_
+#define RDTSC_H_
+
+static inline uint64 rdtsc() {
+	uint64 result;
+#if defined(__x86_64__) || defined(__i386__) || defined(WIN32)
+	result = __rdtsc();
+#else
+	result = 0;
+#endif
+
+	return result;
+}
+
+#endif
-- 
1.8.3.1

From 0694b70d681186ad10af2837560bac8ab1cba8b2 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Thu, 19 Dec 2019 04:28:28 +0000
Subject: [PATCH 1/2] Adding the pg_stat_waitaccum view which shows counts and
 duration of each wait events. Each backend/backgrounds counts and measures
 the time of wait event in every pgstat_report_wait_start and
 pgstat_report_wait_end. They store those info into their local variables and
 send to Statistics Collector. We can get those info via Statistics Collector.

For reducing overhead, I implemented statistic hash instead of
dynamic hash. I also implemented track_wait_timing which
determines wait event duration is collected or not.

On windows, this function might be not worked correctly, because
now it initialize local variables in pg_stat_init which is not
passed to fork processes on windows.
---
 src/backend/catalog/system_views.sql          |   8 +
 src/backend/postmaster/pgstat.c               | 344 ++++++++++++++++++++++++++
 src/backend/storage/lmgr/lwlock.c             |  19 ++
 src/backend/utils/adt/pgstatfuncs.c           |  80 ++++++
 src/backend/utils/misc/guc.c                  |   9 +
 src/backend/utils/misc/postgresql.conf.sample |   1 +
 src/include/catalog/pg_proc.dat               |   9 +
 src/include/pgstat.h                          | 123 ++++++++-
 src/include/storage/lwlock.h                  |   1 +
 src/include/storage/proc.h                    |   1 +
 src/test/regress/expected/rules.out           |   5 +
 11 files changed, 598 insertions(+), 2 deletions(-)

diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index f7800f0..976ad98 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -951,6 +951,14 @@ CREATE VIEW pg_stat_bgwriter AS
         pg_stat_get_buf_alloc() AS buffers_alloc,
         pg_stat_get_bgwriter_stat_reset_time() AS stats_reset;
 
+CREATE VIEW pg_stat_waitaccum AS
+    SELECT
+		S.wait_event_type AS wait_event_type,
+		S.wait_event AS wait_event,
+		S.calls AS calls,
+		S.times AS times
+	FROM pg_stat_get_waitaccum(NULL) AS S;
+
 CREATE VIEW pg_stat_progress_vacuum AS
     SELECT
         S.pid AS pid, S.datid AS datid, D.datname AS datname,
diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index e931512..1454e77 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -123,6 +123,7 @@
  */
 bool		pgstat_track_activities = false;
 bool		pgstat_track_counts = false;
+bool		pgstat_track_wait_timing = false;
 int			pgstat_track_functions = TRACK_FUNC_OFF;
 int			pgstat_track_activity_query_size = 1024;
 
@@ -153,6 +154,10 @@ static time_t last_pgstat_start_time;
 
 static bool pgStatRunningInCollector = false;
 
+WAHash *wa_hash;
+
+instr_time waitStart;
+
 /*
  * Structures in which backends store per-table info that's waiting to be
  * sent to the collector.
@@ -255,6 +260,7 @@ static int	localNumBackends = 0;
  */
 static PgStat_ArchiverStats archiverStats;
 static PgStat_GlobalStats globalStats;
+static PgStat_WaitAccumStats waitAccumStats;
 
 /*
  * List of OIDs of databases we need to write out.  If an entry is InvalidOid,
@@ -280,6 +286,8 @@ static pid_t pgstat_forkexec(void);
 #endif
 
 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
+static void pgstat_init_waitaccum_hash(WAHash **hash);
+static PgStat_WaitAccumEntry *pgstat_add_wa_entry(WAHash *hash, uint32 key);
 static void pgstat_beshutdown_hook(int code, Datum arg);
 
 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
@@ -287,8 +295,11 @@ static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
 												 Oid tableoid, bool create);
 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
+static void pgstat_write_waitaccum_statsfile(FILE *fpout);
 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
+static bool pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+											FILE *fpin, const char *statfile);
 static void backend_read_statsfile(void);
 static void pgstat_read_current_status(void);
 
@@ -324,6 +335,7 @@ static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
+static void pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len);
 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
@@ -331,6 +343,27 @@ static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
 static void pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len);
 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
 
+
+PgStat_WaitAccumEntry *
+pgstat_get_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *current;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	current = hash->buckets[bucket];
+
+	while (current != NULL)
+	{
+		if (current->key == key)
+			return current->entry;
+
+		current = current->next;
+	}
+
+	return NULL;
+}
+
+
 /* ------------------------------------------------------------
  * Public functions called from postmaster follow
  * ------------------------------------------------------------
@@ -602,6 +635,8 @@ retry2:
 
 	pg_freeaddrinfo_all(hints.ai_family, addrs);
 
+	pgstat_init_waitaccum_hash(&wa_hash);
+
 	return;
 
 startup_failed:
@@ -624,6 +659,75 @@ startup_failed:
 	SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
 }
 
+static PgStat_WaitAccumEntry *
+pgstat_add_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *prev;
+	WAEntry *new;
+	int bucket = key % WA_BUCKET_SIZE;
+	
+	prev = hash->buckets[bucket];
+
+	while (prev != NULL && prev->next != NULL)
+		prev = prev->next;
+	
+	new = &hash->entries[hash->entry_num++];
+	new->key = key;
+	new->entry = MemoryContextAllocZero(TopMemoryContext, (sizeof(PgStat_WaitAccumEntry)));
+
+	if (prev != NULL)
+		prev->next = new;
+	else
+		hash->buckets[bucket] = new;
+
+	return new->entry;
+}
+
+static void
+pgstat_init_waitaccum_entry(WAHash *hash, uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+
+	entry = pgstat_add_wa_entry(hash, wait_event_info);
+	entry->wait_event_info = wait_event_info;
+}
+
+static void
+pgstat_init_waitaccum_hash(WAHash **hash)
+{
+	uint32 i;
+	int last_tranche_id;
+
+	*hash = MemoryContextAllocZero(TopMemoryContext, sizeof(WAHash));
+
+	last_tranche_id = LWLockGetLastTrancheId();
+	for (i = PG_WAIT_LWLOCK + 1; i <= (PG_WAIT_LWLOCK | last_tranche_id); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = (PG_WAIT_LOCK | LOCKTAG_RELATION); i <= (PG_WAIT_LOCK | LOCKTAG_LAST_TYPE); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_BUFFER_PIN; i <= PG_WAIT_BUFFER_PIN; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_ACTIVITY; i <= PG_WAIT_ACTIVITY_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_CLIENT; i <= PG_WAIT_CLIENT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	//do extension stuff
+
+	for (i = PG_WAIT_IPC; i <= PG_WAIT_IPC_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_TIMEOUT; i <= PG_WAIT_TIMEOUT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IO; i <= PG_WAIT_IO_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+}
+
 /*
  * subroutine for pgstat_reset_all
  */
@@ -904,6 +1008,9 @@ pgstat_report_stat(bool force)
 
 	/* Now, send function statistics */
 	pgstat_send_funcstats();
+
+	/* Send wait accumulative statistics */
+	pgstat_send_waitaccum();
 }
 
 /*
@@ -1334,6 +1441,8 @@ pgstat_reset_shared_counters(const char *target)
 		msg.m_resettarget = RESET_ARCHIVER;
 	else if (strcmp(target, "bgwriter") == 0)
 		msg.m_resettarget = RESET_BGWRITER;
+	else if (strcmp(target, "waitaccum") == 0)
+		msg.m_resettarget = RESET_WAITACCUM;
 	else
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
@@ -2618,6 +2727,22 @@ pgstat_fetch_global(void)
 	return &globalStats;
 }
 
+/*
+ * ---------
+ * pgstat_fetch_stat_waitaccum() -
+ *
+ *	Support function for the SQL-callable pgstat* functions. Returns
+ *	a pointer to the wait accum statistics struct.
+ * ---------
+ */
+PgStat_WaitAccumStats *
+pgstat_fetch_stat_waitaccum(void)
+{
+	backend_read_statsfile();
+
+	return &waitAccumStats;
+}
+
 
 /* ------------------------------------------------------------
  * Functions for management of the shared-memory PgBackendStatus array
@@ -4407,6 +4532,53 @@ pgstat_send_bgwriter(void)
 	MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
 }
 
+/* ----------
+ * pgstat_send_waitaccum() -
+ *
+ * ----------
+ */
+void
+pgstat_send_waitaccum()
+{
+	PgStat_MsgWaitAccum msg;
+	PgStat_WaitAccumEntry *entry;
+	int i;
+
+	if (wa_hash == NULL)
+		return;
+
+	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_WAITACCUM);
+	msg.m_nentries = 0;
+
+	for (i = 0; i < wa_hash->entry_num; i++)
+	{
+		entry = wa_hash->entries[i].entry;
+
+		/* Send only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Prepare and send the message
+		 */
+		memcpy(&msg.m_entry[msg.m_nentries], entry, sizeof(PgStat_WaitAccumEntry));
+		if (++msg.m_nentries >= PGSTAT_NUM_WAITACCUMENTRIES)
+		{
+			pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+						msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+			msg.m_nentries = 0;
+		}
+
+		/* Clear wait events information. */
+		entry->calls = 0;
+		INSTR_TIME_SET_ZERO(entry->times);
+	}
+
+	if (msg.m_nentries > 0)
+		pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+					msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+}
+
 
 /* ----------
  * PgstatCollectorMain() -
@@ -4599,6 +4771,10 @@ PgstatCollectorMain(int argc, char *argv[])
 					pgstat_recv_bgwriter(&msg.msg_bgwriter, len);
 					break;
 
+				case PGSTAT_MTYPE_WAITACCUM:
+					pgstat_recv_waitaccum(&msg.msg_waitaccum, len);
+					break;
+
 				case PGSTAT_MTYPE_FUNCSTAT:
 					pgstat_recv_funcstat(&msg.msg_funcstat, len);
 					break;
@@ -4869,6 +5045,8 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 	rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
 	(void) rc;					/* we'll check for error with ferror */
 
+	pgstat_write_waitaccum_statsfile(fpout);
+
 	/*
 	 * Walk through the database table.
 	 */
@@ -5074,6 +5252,43 @@ pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
 }
 
 /* ----------
+ * pgstat_write_waitaccum_statsfile() -
+ *		Write the waitAccumStats to the stat file.
+ *
+ * ----------
+ */
+static void
+pgstat_write_waitaccum_statsfile(FILE *fpout)
+{
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			rc;
+	int			i;
+
+	/*
+	 * Walk through the waitaccum hash.
+	 */
+	for (i = 0; i < hash->entry_num; i++)
+	{
+		entry = hash->entries[i].entry;
+
+		/* Write only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Write out the DB entry. We don't write the tables or functions
+		 * pointers, since they're of no use to any other process.
+		 */
+		fputc('D', fpout);
+		rc = fwrite(entry, sizeof(PgStat_WaitAccumEntry), 1, fpout);
+		(void) rc;				/* we'll check for error with ferror */
+	}	
+
+	fputc('E', fpout);
+}
+
+/* ----------
  * pgstat_read_statsfiles() -
  *
  *	Reads in some existing statistics collector files and returns the
@@ -5126,6 +5341,7 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 	 */
 	memset(&globalStats, 0, sizeof(globalStats));
 	memset(&archiverStats, 0, sizeof(archiverStats));
+	waitAccumStats.hash = MemoryContextAllocZero(pgStatLocalContext, sizeof(WAHash));
 
 	/*
 	 * Set the current timestamp (will be kept only in case we can't load an
@@ -5196,6 +5412,9 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 		goto done;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&waitAccumStats, fpin, statfile))
+		goto done;
+
 	/*
 	 * We found an existing collector stats file. Read it and put all the
 	 * hashtable entries into place.
@@ -5494,10 +5713,13 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 	PgStat_StatDBEntry dbentry;
 	PgStat_GlobalStats myGlobalStats;
 	PgStat_ArchiverStats myArchiverStats;
+	PgStat_WaitAccumStats myWaitAccumStats;
 	FILE	   *fpin;
 	int32		format_id;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
 
+	myWaitAccumStats.hash = MemoryContextAllocZero(CurrentMemoryContext, sizeof(WAHash));
+
 	/*
 	 * Try to open the stats file.  As above, anything but ENOENT is worthy of
 	 * complaining about.
@@ -5548,6 +5770,9 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 		return false;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&myWaitAccumStats, fpin, statfile))
+		return false;
+
 	/* By default, we're going to return the timestamp of the global file. */
 	*ts = myGlobalStats.stats_timestamp;
 
@@ -5601,6 +5826,75 @@ done:
 	return true;
 }
 
+/* ----------
+ * pgstat_read_statsfiles() -
+ *
+ *	Reads the waitaccum stats from the file.
+ *	If an error happens when reading file, return false. Otherwise return true.
+ *
+ * ----------
+ */
+static bool
+pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+								FILE *fpin, const char *statfile)
+{
+	PgStat_WaitAccumEntry *entry;
+	PgStat_WaitAccumEntry buf;
+	WAHash *hash = stats->hash;
+
+	/*
+	 * Read and put all the hashtable entries into place.
+	 */
+	for (;;)
+	{
+		switch (fgetc(fpin))
+		{
+				/*
+				 * 'D'	A PgStat_WaitAccumEntry struct describing a database
+				 * follows.
+				 */
+			case 'D':
+				if (fread(&buf, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+									 != sizeof(PgStat_WaitAccumEntry))
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				entry = pgstat_get_wa_entry(hash, buf.wait_event_info);
+
+				if (entry)
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				/*
+				 * Add to the DB hash
+				 */
+				entry = pgstat_add_wa_entry(hash, buf.wait_event_info);
+				memcpy(entry, &buf, sizeof(PgStat_WaitAccumEntry));
+
+				break;
+
+			case 'E':
+				return true;
+
+			default:
+				ereport(pgStatRunningInCollector ? LOG : WARNING,
+						(errmsg("corrupted statistics file \"%s\"",
+								statfile)));
+				return false;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * If not already done, read the statistics collector stats file into
  * some hash tables.  The results will be kept until pgstat_clear_snapshot()
@@ -6110,7 +6404,20 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 		memset(&archiverStats, 0, sizeof(archiverStats));
 		archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
 	}
+	else if (msg->m_resettarget == RESET_WAITACCUM)
+	{
+		PgStat_WaitAccumEntry *entry;
+		WAHash *hash = waitAccumStats.hash;
+		int i;
+
+		for (i = 0; i < hash->entry_num; i++)
+		{
+			entry = hash->entries[i].entry;
 
+			entry->calls = 0;
+			INSTR_TIME_SET_ZERO(entry->times);
+		}
+	}
 	/*
 	 * Presumably the sender of this message validated the target, don't
 	 * complain here if it's not valid
@@ -6290,6 +6597,43 @@ pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
 }
 
 /* ----------
+ * pgstat_recv_waitaccum() -
+ *
+ *	Process a WAITACCUM message.
+ * ----------
+ */
+static void
+pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
+{
+	PgStat_WaitAccumEntry *m_entry = &(msg->m_entry[0]);
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			i;
+
+	/*
+	 * Process all function entries in the message.
+	 */
+	for (i = 0; i < msg->m_nentries; i++, m_entry++)
+	{
+		entry = pgstat_get_wa_entry(hash, m_entry->wait_event_info);
+
+		if (!entry)
+		{
+			entry = pgstat_add_wa_entry(hash, m_entry->wait_event_info);
+			memcpy(entry, m_entry, sizeof(PgStat_WaitAccumEntry));
+		}
+		else
+		{
+			/*
+			 * Otherwise add the values to the existing entry.
+			 */
+			entry->calls += m_entry->calls;
+			INSTR_TIME_ADD(entry->times, m_entry->times);
+		}
+	}
+}
+
+/* ----------
  * pgstat_recv_recoveryconflict() -
  *
  *	Process a RECOVERYCONFLICT message.
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 18e3843..8f5b0ba 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -594,6 +594,25 @@ LWLockNewTrancheId(void)
 }
 
 /*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
+/*
  * Register a tranche ID in the lookup table for the current process.  This
  * routine will save a pointer to the tranche name passed as an argument,
  * so the name should be allocated in a backend-lifetime context
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 05240bf..b408db3 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -1970,3 +1970,83 @@ pg_stat_get_archiver(PG_FUNCTION_ARGS)
 	PG_RETURN_DATUM(HeapTupleGetDatum(
 									  heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	PgStat_WaitAccumStats *waitaccum_stats;
+	PgStat_WaitAccumEntry *entry;
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	/* Get statistics about the waitaccum process */
+	waitaccum_stats = pgstat_fetch_stat_waitaccum();
+
+	for (i = 0; i < waitaccum_stats->hash->entry_num; i++)
+	{
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		entry = waitaccum_stats->hash->entries[i].entry;
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 8d951ce..ebae427 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1424,6 +1424,15 @@ static struct config_bool ConfigureNamesBool[] =
 		false,
 		NULL, NULL, NULL
 	},
+	{
+		{"track_wait_timing", PGC_SUSET, STATS_COLLECTOR,
+			gettext_noop("Collects timing statistics for wait events."),
+			NULL
+		},
+		&pgstat_track_wait_timing,
+		false,
+		NULL, NULL, NULL
+	},
 
 	{
 		{"update_process_title", PGC_SUSET, PROCESS_TITLE,
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index 087190c..070c213 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -569,6 +569,7 @@
 #track_activities = on
 #track_counts = on
 #track_io_timing = off
+#track_wait_timing = off
 #track_functions = none			# none, pl, all
 #track_activity_query_size = 1024	# (change requires restart)
 #stats_temp_directory = 'pg_stat_tmp'
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index ac8f64b..9d7c2e8 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5152,6 +5152,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,sslcompression,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '2228',
+  descr => 'statistics: information about accumulative data of wait event',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 's', proparallel => 'r',
+  prorettype => 'record', proargtypes => 'int4',
+  proallargtypes => '{int4,text,text,int8,int8}',
+  proargmodes => '{i,o,o,o,o}',
+  proargnames => '{pid,wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index fe076d8..32c4b5f 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -59,6 +59,7 @@ typedef enum StatMsgType
 	PGSTAT_MTYPE_ANALYZE,
 	PGSTAT_MTYPE_ARCHIVER,
 	PGSTAT_MTYPE_BGWRITER,
+	PGSTAT_MTYPE_WAITACCUM,
 	PGSTAT_MTYPE_FUNCSTAT,
 	PGSTAT_MTYPE_FUNCPURGE,
 	PGSTAT_MTYPE_RECOVERYCONFLICT,
@@ -119,7 +120,8 @@ typedef struct PgStat_TableCounts
 typedef enum PgStat_Shared_Reset_Target
 {
 	RESET_ARCHIVER,
-	RESET_BGWRITER
+	RESET_BGWRITER,
+	RESET_WAITACCUM
 } PgStat_Shared_Reset_Target;
 
 /* Possible object types for resetting single counters */
@@ -423,6 +425,33 @@ typedef struct PgStat_MsgBgWriter
 } PgStat_MsgBgWriter;
 
 /* ----------
+ * PgStat_WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct PgStat_WaitAccumEntry
+{
+	uint32			wait_event_info;
+	PgStat_Counter	calls;
+	instr_time		times;
+} PgStat_WaitAccumEntry;
+
+/* ----------
+ * PgStat_MsgWaitAccum	Sent by backend/background's process to update statistics.
+ * ----------
+ */
+#define PGSTAT_NUM_WAITACCUMENTRIES	\
+	((PGSTAT_MSG_PAYLOAD - sizeof(int))  \
+	 / sizeof(PgStat_WaitAccumEntry))
+
+typedef struct PgStat_MsgWaitAccum
+{
+	PgStat_MsgHdr m_hdr;
+
+	int m_nentries;
+	PgStat_WaitAccumEntry m_entry[PGSTAT_NUM_WAITACCUMENTRIES];
+} PgStat_MsgWaitAccum;
+
+/* ----------
  * PgStat_MsgRecoveryConflict	Sent by the backend upon recovery conflict
  * ----------
  */
@@ -564,6 +593,7 @@ typedef union PgStat_Msg
 	PgStat_MsgAnalyze msg_analyze;
 	PgStat_MsgArchiver msg_archiver;
 	PgStat_MsgBgWriter msg_bgwriter;
+	PgStat_MsgWaitAccum msg_waitaccum;
 	PgStat_MsgFuncstat msg_funcstat;
 	PgStat_MsgFuncpurge msg_funcpurge;
 	PgStat_MsgRecoveryConflict msg_recoveryconflict;
@@ -581,7 +611,7 @@ typedef union PgStat_Msg
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9D
+#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9E
 
 /* ----------
  * PgStat_StatDBEntry			The collector's data per database
@@ -711,6 +741,30 @@ typedef struct PgStat_GlobalStats
 	TimestampTz stat_reset_timestamp;
 } PgStat_GlobalStats;
 
+typedef struct WAEntry
+{
+	int key;
+	PgStat_WaitAccumEntry *entry;
+	struct WAEntry *next;
+} WAEntry;
+
+#define WA_BUCKET_SIZE 461
+
+typedef struct WAHash
+{
+	WAEntry entries[WA_BUCKET_SIZE];
+	WAEntry *buckets[WA_BUCKET_SIZE];
+	int entry_num;
+} WAHash;
+
+/*
+ * WaitAccum statistics kept in the stats collector
+ */
+typedef struct PgStat_WaitAccumStats
+{
+	WAHash *hash;
+} PgStat_WaitAccumStats;
+
 
 /* ----------
  * Backend types
@@ -787,6 +841,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+
 /* ----------
  * Wait Events - Client
  *
@@ -808,6 +864,8 @@ typedef enum
 	WAIT_EVENT_GSS_OPEN_SERVER,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_GSS_OPEN_SERVER
+
 /* ----------
  * Wait Events - IPC
  *
@@ -856,6 +914,8 @@ typedef enum
 	WAIT_EVENT_SYNC_REP
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_SYNC_REP
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -869,6 +929,8 @@ typedef enum
 	WAIT_EVENT_RECOVERY_APPLY_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_RECOVERY_APPLY_DELAY
+
 /* ----------
  * Wait Events - IO
  *
@@ -947,6 +1009,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_WAL_WRITE
+
 /* ----------
  * Command type for progress reporting purposes
  * ----------
@@ -1203,6 +1267,8 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
+extern WAHash *wa_hash;
+extern instr_time waitStart;
 
 /* ----------
  * GUC parameters
@@ -1210,6 +1276,7 @@ typedef struct PgStat_FunctionCallUsage
  */
 extern bool pgstat_track_activities;
 extern bool pgstat_track_counts;
+extern bool pgstat_track_wait_timing;
 extern int	pgstat_track_functions;
 extern PGDLLIMPORT int pgstat_track_activity_query_size;
 extern char *pgstat_stat_directory;
@@ -1227,6 +1294,7 @@ extern PgStat_MsgBgWriter BgWriterStats;
 extern PgStat_Counter pgStatBlockReadTime;
 extern PgStat_Counter pgStatBlockWriteTime;
 
+extern PgStat_WaitAccumEntry *pgstat_get_wa_entry(WAHash *hash, uint32 key);
 /* ----------
  * Functions called from postmaster
  * ----------
@@ -1314,6 +1382,50 @@ extern char *pgstat_clip_activity(const char *raw_activity);
  * initialized.
  * ----------
  */
+
+static inline void
+pgstat_report_waitaccum_start()
+{
+	if (wa_hash == NULL)
+		return; 
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(waitStart);
+	}
+}
+
+static inline void
+pgstat_report_waitaccum_end(uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+	instr_time  diff;
+
+	if (wa_hash == NULL)
+		return; 
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(diff);
+		INSTR_TIME_SUBTRACT(diff, waitStart);
+	}
+
+	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
+
+	if (!entry)
+	{
+		printf("wait_event_info: %u.\n", wait_event_info);
+		fflush(stdout);
+		return;
+	}
+
+	entry->calls++;
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+}
+
 static inline void
 pgstat_report_wait_start(uint32 wait_event_info)
 {
@@ -1327,6 +1439,8 @@ pgstat_report_wait_start(uint32 wait_event_info)
 	 * four-bytes, updates are atomic.
 	 */
 	proc->wait_event_info = wait_event_info;
+
+	pgstat_report_waitaccum_start();
 }
 
 /* ----------
@@ -1346,6 +1460,8 @@ pgstat_report_wait_end(void)
 	if (!pgstat_track_activities || !proc)
 		return;
 
+	pgstat_report_waitaccum_end(proc->wait_event_info);
+
 	/*
 	 * Since this is a four-byte field which is always read and written as
 	 * four-bytes, updates are atomic.
@@ -1353,6 +1469,7 @@ pgstat_report_wait_end(void)
 	proc->wait_event_info = 0;
 }
 
+
 /* nontransactional event counts are simple enough to inline */
 
 #define pgstat_count_heap_scan(rel)									\
@@ -1420,6 +1537,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 
 extern void pgstat_send_archiver(const char *xlog, bool failed);
 extern void pgstat_send_bgwriter(void);
+extern void pgstat_send_waitaccum(void);
 
 /* ----------
  * Support functions for the SQL-callable functions to
@@ -1434,5 +1552,6 @@ extern PgStat_StatFuncEntry *pgstat_fetch_stat_funcentry(Oid funcid);
 extern int	pgstat_fetch_stat_numbackends(void);
 extern PgStat_ArchiverStats *pgstat_fetch_stat_archiver(void);
 extern PgStat_GlobalStats *pgstat_fetch_global(void);
+extern PgStat_WaitAccumStats *pgstat_fetch_stat_waitaccum(void);
 
 #endif							/* PGSTAT_H */
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index f9450da..cf2e953 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -184,6 +184,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 281e1db..4d03d80 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -21,6 +21,7 @@
 #include "storage/lock.h"
 #include "storage/pg_sema.h"
 #include "storage/proclist_types.h"
+#include "portability/instr_time.h"
 
 /*
  * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 80a0782..b524fae 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2058,6 +2058,11 @@ pg_stat_user_tables| SELECT pg_stat_all_tables.relid,
     pg_stat_all_tables.autoanalyze_count
    FROM pg_stat_all_tables
   WHERE ((pg_stat_all_tables.schemaname <> ALL (ARRAY['pg_catalog'::name, 'information_schema'::name])) AND (pg_stat_all_tables.schemaname !~ '^pg_toast'::text));
+pg_stat_waitaccum| SELECT s.wait_event_type,
+    s.wait_event,
+    s.calls,
+    s.times
+   FROM pg_stat_get_waitaccum(NULL::integer) s(wait_event_type, wait_event, calls, times);
 pg_stat_wal_receiver| SELECT s.pid,
     s.status,
     s.receive_start_lsn,
-- 
1.8.3.1

From 00cac6dce8b8a533e1fad33035a41aafc79d657c Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Thu, 19 Dec 2019 04:47:16 +0000
Subject: [PATCH 2/2] [POC] Changed measuring method of wait event timed from
 INSTR_TIME (which uses gettimeofday or clock_gettime) to rdtsc. This might
 reduce the overhead of measuring overhead.

Any supports like changing clock cycle to actual time or error
handling are not currently implemented.
---
 src/backend/postmaster/pgstat.c      |  8 ++++----
 src/backend/utils/adt/pgstatfuncs.c  |  2 +-
 src/include/pgstat.h                 | 14 +++++++-------
 src/include/portability/instr_time.h | 23 +++++++++++++++++++++++
 4 files changed, 35 insertions(+), 12 deletions(-)

diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 1454e77..f9d72e7 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -156,7 +156,7 @@ static bool pgStatRunningInCollector = false;
 
 WAHash *wa_hash;
 
-instr_time waitStart;
+uint64 waitStart;
 
 /*
  * Structures in which backends store per-table info that's waiting to be
@@ -4571,7 +4571,7 @@ pgstat_send_waitaccum()
 
 		/* Clear wait events information. */
 		entry->calls = 0;
-		INSTR_TIME_SET_ZERO(entry->times);
+		entry->times = 0;
 	}
 
 	if (msg.m_nentries > 0)
@@ -6415,7 +6415,7 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 			entry = hash->entries[i].entry;
 
 			entry->calls = 0;
-			INSTR_TIME_SET_ZERO(entry->times);
+			entry->times = 0;
 		}
 	}
 	/*
@@ -6628,7 +6628,7 @@ pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
 			 * Otherwise add the values to the existing entry.
 			 */
 			entry->calls += m_entry->calls;
-			INSTR_TIME_ADD(entry->times, m_entry->times);
+			entry->times += m_entry->times;
 		}
 	}
 }
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index b408db3..f759c7d 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -2040,7 +2040,7 @@ pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
 
 		values[2] = Int64GetDatum(entry->calls);
 
-		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+		values[3] = UInt64GetDatum(entry->times);
 
 		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
 	}
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 32c4b5f..2515893 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -432,7 +432,7 @@ typedef struct PgStat_WaitAccumEntry
 {
 	uint32			wait_event_info;
 	PgStat_Counter	calls;
-	instr_time		times;
+	uint64			times;
 } PgStat_WaitAccumEntry;
 
 /* ----------
@@ -1268,7 +1268,7 @@ typedef struct PgStat_FunctionCallUsage
 } PgStat_FunctionCallUsage;
 
 extern WAHash *wa_hash;
-extern instr_time waitStart;
+extern uint64 waitStart;
 
 /* ----------
  * GUC parameters
@@ -1391,7 +1391,7 @@ pgstat_report_waitaccum_start()
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(waitStart);
+		waitStart = rdtsc();
 	}
 }
 
@@ -1399,15 +1399,15 @@ static inline void
 pgstat_report_waitaccum_end(uint32 wait_event_info)
 {
 	PgStat_WaitAccumEntry *entry;
-	instr_time  diff;
+	uint64		diff = 0;
 
 	if (wa_hash == NULL)
 		return; 
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(diff);
-		INSTR_TIME_SUBTRACT(diff, waitStart);
+		diff = rdtsc();
+		diff -= waitStart;
 	}
 
 	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
@@ -1422,7 +1422,7 @@ pgstat_report_waitaccum_end(uint32 wait_event_info)
 	entry->calls++;
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_ADD(entry->times, diff);
+		entry->times += diff;
 	}
 }
 
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 0f5c161..668fa63 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -57,6 +57,10 @@
 
 #ifndef WIN32
 
+#if defined(__x86_64__) || defined(__i386__)
+#include <x86intrin.h>
+#endif
+
 #ifdef HAVE_CLOCK_GETTIME
 
 /* Use clock_gettime() */
@@ -209,6 +213,8 @@ typedef struct timeval instr_time;
 
 #else							/* WIN32 */
 
+#include <intrin.h>
+
 /* Use QueryPerformanceCounter() */
 
 typedef LARGE_INTEGER instr_time;
@@ -254,3 +260,20 @@ GetTimerFrequency(void)
 	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
 
 #endif							/* INSTR_TIME_H */
+
+
+#ifndef RDTSC_H_
+#define RDTSC_H_
+
+static inline uint64 rdtsc() {
+	uint64 result;
+#if defined(__x86_64__) || defined(__i386__) || defined(WIN32)
+	result = __rdtsc();
+#else
+	result = 0;
+#endif
+
+	return result;
+}
+
+#endif
-- 
1.8.3.1

From b009b1f8f6be47ae61b5e4538e2730d721ee60db Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Wed, 15 Jan 2020 09:13:19 +0000
Subject: [PATCH v4 1/2] Add pg_stat_waitaccum view.

pg_stat_waitaccum shows counts and duration of each wait events.
Each backend/backgrounds counts and measures the time of wait event
in every pgstat_report_wait_start and pgstat_report_wait_end. They
store those info into their local variables and send to Statistics
Collector. We can get those info via Statistics Collector.

For reducing overhead, I implemented statistic hash instead of
dynamic hash. I also implemented track_wait_timing which
determines wait event duration is collected or not.

On windows, this function might be not worked correctly, because
now it initializes local variables in pg_stat_init which is not
passed to fork processes on windows.
---
 src/backend/catalog/system_views.sql          |   8 +
 src/backend/postmaster/pgstat.c               | 344 ++++++++++++++++++++++++++
 src/backend/storage/lmgr/lwlock.c             |  19 ++
 src/backend/utils/adt/pgstatfuncs.c           |  80 ++++++
 src/backend/utils/misc/guc.c                  |   9 +
 src/backend/utils/misc/postgresql.conf.sample |   1 +
 src/include/catalog/pg_proc.dat               |   9 +
 src/include/pgstat.h                          | 123 ++++++++-
 src/include/storage/lwlock.h                  |   1 +
 src/include/storage/proc.h                    |   1 +
 src/test/regress/expected/rules.out           |   5 +
 11 files changed, 598 insertions(+), 2 deletions(-)

diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index 773edf8..80f2caa 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -957,6 +957,14 @@ CREATE VIEW pg_stat_bgwriter AS
         pg_stat_get_buf_alloc() AS buffers_alloc,
         pg_stat_get_bgwriter_stat_reset_time() AS stats_reset;
 
+CREATE VIEW pg_stat_waitaccum AS
+    SELECT
+		S.wait_event_type AS wait_event_type,
+		S.wait_event AS wait_event,
+		S.calls AS calls,
+		S.times AS times
+	FROM pg_stat_get_waitaccum(NULL) AS S;
+
 CREATE VIEW pg_stat_progress_vacuum AS
     SELECT
         S.pid AS pid, S.datid AS datid, D.datname AS datname,
diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 51c486b..08e10ad 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -123,6 +123,7 @@
  */
 bool		pgstat_track_activities = false;
 bool		pgstat_track_counts = false;
+bool		pgstat_track_wait_timing = false;
 int			pgstat_track_functions = TRACK_FUNC_OFF;
 int			pgstat_track_activity_query_size = 1024;
 
@@ -153,6 +154,10 @@ static time_t last_pgstat_start_time;
 
 static bool pgStatRunningInCollector = false;
 
+WAHash *wa_hash;
+
+instr_time waitStart;
+
 /*
  * Structures in which backends store per-table info that's waiting to be
  * sent to the collector.
@@ -255,6 +260,7 @@ static int	localNumBackends = 0;
  */
 static PgStat_ArchiverStats archiverStats;
 static PgStat_GlobalStats globalStats;
+static PgStat_WaitAccumStats waitAccumStats;
 
 /*
  * List of OIDs of databases we need to write out.  If an entry is InvalidOid,
@@ -280,6 +286,8 @@ static pid_t pgstat_forkexec(void);
 #endif
 
 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
+static void pgstat_init_waitaccum_hash(WAHash **hash);
+static PgStat_WaitAccumEntry *pgstat_add_wa_entry(WAHash *hash, uint32 key);
 static void pgstat_beshutdown_hook(int code, Datum arg);
 
 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
@@ -287,8 +295,11 @@ static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
 												 Oid tableoid, bool create);
 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
+static void pgstat_write_waitaccum_statsfile(FILE *fpout);
 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
+static bool pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+											FILE *fpin, const char *statfile);
 static void backend_read_statsfile(void);
 static void pgstat_read_current_status(void);
 
@@ -324,6 +335,7 @@ static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
+static void pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len);
 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
@@ -331,6 +343,27 @@ static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
 static void pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len);
 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
 
+
+PgStat_WaitAccumEntry *
+pgstat_get_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *current;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	current = hash->buckets[bucket];
+
+	while (current != NULL)
+	{
+		if (current->key == key)
+			return current->entry;
+
+		current = current->next;
+	}
+
+	return NULL;
+}
+
+
 /* ------------------------------------------------------------
  * Public functions called from postmaster follow
  * ------------------------------------------------------------
@@ -602,6 +635,8 @@ retry2:
 
 	pg_freeaddrinfo_all(hints.ai_family, addrs);
 
+	pgstat_init_waitaccum_hash(&wa_hash);
+
 	return;
 
 startup_failed:
@@ -624,6 +659,75 @@ startup_failed:
 	SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
 }
 
+static PgStat_WaitAccumEntry *
+pgstat_add_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *prev;
+	WAEntry *new;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	prev = hash->buckets[bucket];
+
+	while (prev != NULL && prev->next != NULL)
+		prev = prev->next;
+
+	new = &hash->entries[hash->entry_num++];
+	new->key = key;
+	new->entry = MemoryContextAllocZero(TopMemoryContext, (sizeof(PgStat_WaitAccumEntry)));
+
+	if (prev != NULL)
+		prev->next = new;
+	else
+		hash->buckets[bucket] = new;
+
+	return new->entry;
+}
+
+static void
+pgstat_init_waitaccum_entry(WAHash *hash, uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+
+	entry = pgstat_add_wa_entry(hash, wait_event_info);
+	entry->wait_event_info = wait_event_info;
+}
+
+static void
+pgstat_init_waitaccum_hash(WAHash **hash)
+{
+	uint32 i;
+	int last_tranche_id;
+
+	*hash = MemoryContextAllocZero(TopMemoryContext, sizeof(WAHash));
+
+	last_tranche_id = LWLockGetLastTrancheId();
+	for (i = PG_WAIT_LWLOCK + 1; i <= (PG_WAIT_LWLOCK | last_tranche_id); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = (PG_WAIT_LOCK | LOCKTAG_RELATION); i <= (PG_WAIT_LOCK | LOCKTAG_LAST_TYPE); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_BUFFER_PIN; i <= PG_WAIT_BUFFER_PIN; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_ACTIVITY; i <= PG_WAIT_ACTIVITY_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_CLIENT; i <= PG_WAIT_CLIENT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	//do extension stuff
+
+	for (i = PG_WAIT_IPC; i <= PG_WAIT_IPC_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_TIMEOUT; i <= PG_WAIT_TIMEOUT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IO; i <= PG_WAIT_IO_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+}
+
 /*
  * subroutine for pgstat_reset_all
  */
@@ -904,6 +1008,9 @@ pgstat_report_stat(bool force)
 
 	/* Now, send function statistics */
 	pgstat_send_funcstats();
+
+	/* Send wait accumulative statistics */
+	pgstat_send_waitaccum();
 }
 
 /*
@@ -1334,6 +1441,8 @@ pgstat_reset_shared_counters(const char *target)
 		msg.m_resettarget = RESET_ARCHIVER;
 	else if (strcmp(target, "bgwriter") == 0)
 		msg.m_resettarget = RESET_BGWRITER;
+	else if (strcmp(target, "waitaccum") == 0)
+		msg.m_resettarget = RESET_WAITACCUM;
 	else
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
@@ -2618,6 +2727,22 @@ pgstat_fetch_global(void)
 	return &globalStats;
 }
 
+/*
+ * ---------
+ * pgstat_fetch_stat_waitaccum() -
+ *
+ *	Support function for the SQL-callable pgstat* functions. Returns
+ *	a pointer to the wait accum statistics struct.
+ * ---------
+ */
+PgStat_WaitAccumStats *
+pgstat_fetch_stat_waitaccum(void)
+{
+	backend_read_statsfile();
+
+	return &waitAccumStats;
+}
+
 
 /* ------------------------------------------------------------
  * Functions for management of the shared-memory PgBackendStatus array
@@ -4410,6 +4535,53 @@ pgstat_send_bgwriter(void)
 	MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
 }
 
+/* ----------
+ * pgstat_send_waitaccum() -
+ *
+ * ----------
+ */
+void
+pgstat_send_waitaccum()
+{
+	PgStat_MsgWaitAccum msg;
+	PgStat_WaitAccumEntry *entry;
+	int i;
+
+	if (wa_hash == NULL)
+		return;
+
+	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_WAITACCUM);
+	msg.m_nentries = 0;
+
+	for (i = 0; i < wa_hash->entry_num; i++)
+	{
+		entry = wa_hash->entries[i].entry;
+
+		/* Send only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Prepare and send the message
+		 */
+		memcpy(&msg.m_entry[msg.m_nentries], entry, sizeof(PgStat_WaitAccumEntry));
+		if (++msg.m_nentries >= PGSTAT_NUM_WAITACCUMENTRIES)
+		{
+			pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+						msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+			msg.m_nentries = 0;
+		}
+
+		/* Clear wait events information. */
+		entry->calls = 0;
+		INSTR_TIME_SET_ZERO(entry->times);
+	}
+
+	if (msg.m_nentries > 0)
+		pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+					msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+}
+
 
 /* ----------
  * PgstatCollectorMain() -
@@ -4602,6 +4774,10 @@ PgstatCollectorMain(int argc, char *argv[])
 					pgstat_recv_bgwriter(&msg.msg_bgwriter, len);
 					break;
 
+				case PGSTAT_MTYPE_WAITACCUM:
+					pgstat_recv_waitaccum(&msg.msg_waitaccum, len);
+					break;
+
 				case PGSTAT_MTYPE_FUNCSTAT:
 					pgstat_recv_funcstat(&msg.msg_funcstat, len);
 					break;
@@ -4872,6 +5048,8 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 	rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
 	(void) rc;					/* we'll check for error with ferror */
 
+	pgstat_write_waitaccum_statsfile(fpout);
+
 	/*
 	 * Walk through the database table.
 	 */
@@ -5077,6 +5255,43 @@ pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
 }
 
 /* ----------
+ * pgstat_write_waitaccum_statsfile() -
+ *		Write the waitAccumStats to the stat file.
+ *
+ * ----------
+ */
+static void
+pgstat_write_waitaccum_statsfile(FILE *fpout)
+{
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			rc;
+	int			i;
+
+	/*
+	 * Walk through the waitaccum hash.
+	 */
+	for (i = 0; i < hash->entry_num; i++)
+	{
+		entry = hash->entries[i].entry;
+
+		/* Write only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Write out the DB entry. We don't write the tables or functions
+		 * pointers, since they're of no use to any other process.
+		 */
+		fputc('D', fpout);
+		rc = fwrite(entry, sizeof(PgStat_WaitAccumEntry), 1, fpout);
+		(void) rc;				/* we'll check for error with ferror */
+	}
+
+	fputc('E', fpout);
+}
+
+/* ----------
  * pgstat_read_statsfiles() -
  *
  *	Reads in some existing statistics collector files and returns the
@@ -5129,6 +5344,7 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 	 */
 	memset(&globalStats, 0, sizeof(globalStats));
 	memset(&archiverStats, 0, sizeof(archiverStats));
+	waitAccumStats.hash = MemoryContextAllocZero(pgStatLocalContext, sizeof(WAHash));
 
 	/*
 	 * Set the current timestamp (will be kept only in case we can't load an
@@ -5199,6 +5415,9 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 		goto done;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&waitAccumStats, fpin, statfile))
+		goto done;
+
 	/*
 	 * We found an existing collector stats file. Read it and put all the
 	 * hashtable entries into place.
@@ -5497,10 +5716,13 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 	PgStat_StatDBEntry dbentry;
 	PgStat_GlobalStats myGlobalStats;
 	PgStat_ArchiverStats myArchiverStats;
+	PgStat_WaitAccumStats myWaitAccumStats;
 	FILE	   *fpin;
 	int32		format_id;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
 
+	myWaitAccumStats.hash = MemoryContextAllocZero(CurrentMemoryContext, sizeof(WAHash));
+
 	/*
 	 * Try to open the stats file.  As above, anything but ENOENT is worthy of
 	 * complaining about.
@@ -5551,6 +5773,9 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 		return false;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&myWaitAccumStats, fpin, statfile))
+		return false;
+
 	/* By default, we're going to return the timestamp of the global file. */
 	*ts = myGlobalStats.stats_timestamp;
 
@@ -5604,6 +5829,75 @@ done:
 	return true;
 }
 
+/* ----------
+ * pgstat_read_statsfiles() -
+ *
+ *	Reads the waitaccum stats from the file.
+ *	If an error happens when reading file, return false. Otherwise return true.
+ *
+ * ----------
+ */
+static bool
+pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+								FILE *fpin, const char *statfile)
+{
+	PgStat_WaitAccumEntry *entry;
+	PgStat_WaitAccumEntry buf;
+	WAHash *hash = stats->hash;
+
+	/*
+	 * Read and put all the hashtable entries into place.
+	 */
+	for (;;)
+	{
+		switch (fgetc(fpin))
+		{
+				/*
+				 * 'D'	A PgStat_WaitAccumEntry struct describing a database
+				 * follows.
+				 */
+			case 'D':
+				if (fread(&buf, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+									 != sizeof(PgStat_WaitAccumEntry))
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				entry = pgstat_get_wa_entry(hash, buf.wait_event_info);
+
+				if (entry)
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				/*
+				 * Add to the DB hash
+				 */
+				entry = pgstat_add_wa_entry(hash, buf.wait_event_info);
+				memcpy(entry, &buf, sizeof(PgStat_WaitAccumEntry));
+
+				break;
+
+			case 'E':
+				return true;
+
+			default:
+				ereport(pgStatRunningInCollector ? LOG : WARNING,
+						(errmsg("corrupted statistics file \"%s\"",
+								statfile)));
+				return false;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * If not already done, read the statistics collector stats file into
  * some hash tables.  The results will be kept until pgstat_clear_snapshot()
@@ -6113,7 +6407,20 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 		memset(&archiverStats, 0, sizeof(archiverStats));
 		archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
 	}
+	else if (msg->m_resettarget == RESET_WAITACCUM)
+	{
+		PgStat_WaitAccumEntry *entry;
+		WAHash *hash = waitAccumStats.hash;
+		int i;
+
+		for (i = 0; i < hash->entry_num; i++)
+		{
+			entry = hash->entries[i].entry;
 
+			entry->calls = 0;
+			INSTR_TIME_SET_ZERO(entry->times);
+		}
+	}
 	/*
 	 * Presumably the sender of this message validated the target, don't
 	 * complain here if it's not valid
@@ -6293,6 +6600,43 @@ pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
 }
 
 /* ----------
+ * pgstat_recv_waitaccum() -
+ *
+ *	Process a WAITACCUM message.
+ * ----------
+ */
+static void
+pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
+{
+	PgStat_WaitAccumEntry *m_entry = &(msg->m_entry[0]);
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			i;
+
+	/*
+	 * Process all function entries in the message.
+	 */
+	for (i = 0; i < msg->m_nentries; i++, m_entry++)
+	{
+		entry = pgstat_get_wa_entry(hash, m_entry->wait_event_info);
+
+		if (!entry)
+		{
+			entry = pgstat_add_wa_entry(hash, m_entry->wait_event_info);
+			memcpy(entry, m_entry, sizeof(PgStat_WaitAccumEntry));
+		}
+		else
+		{
+			/*
+			 * Otherwise add the values to the existing entry.
+			 */
+			entry->calls += m_entry->calls;
+			INSTR_TIME_ADD(entry->times, m_entry->times);
+		}
+	}
+}
+
+/* ----------
  * pgstat_recv_recoveryconflict() -
  *
  *	Process a RECOVERYCONFLICT message.
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index d07ce60..6f4eb19 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -594,6 +594,25 @@ LWLockNewTrancheId(void)
 }
 
 /*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
+/*
  * Register a tranche ID in the lookup table for the current process.  This
  * routine will save a pointer to the tranche name passed as an argument,
  * so the name should be allocated in a backend-lifetime context
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 3dbf604..bed7d01 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -1974,3 +1974,83 @@ pg_stat_get_archiver(PG_FUNCTION_ARGS)
 	PG_RETURN_DATUM(HeapTupleGetDatum(
 									  heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	PgStat_WaitAccumStats *waitaccum_stats;
+	PgStat_WaitAccumEntry *entry;
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	/* Get statistics about the waitaccum process */
+	waitaccum_stats = pgstat_fetch_stat_waitaccum();
+
+	for (i = 0; i < waitaccum_stats->hash->entry_num; i++)
+	{
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		entry = waitaccum_stats->hash->entries[i].entry;
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index e5f8a13..2924472 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1424,6 +1424,15 @@ static struct config_bool ConfigureNamesBool[] =
 		false,
 		NULL, NULL, NULL
 	},
+	{
+		{"track_wait_timing", PGC_SUSET, STATS_COLLECTOR,
+			gettext_noop("Collects timing statistics for wait events."),
+			NULL
+		},
+		&pgstat_track_wait_timing,
+		false,
+		NULL, NULL, NULL
+	},
 
 	{
 		{"update_process_title", PGC_SUSET, PROCESS_TITLE,
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index e1048c0..3a99182 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -570,6 +570,7 @@
 #track_activities = on
 #track_counts = on
 #track_io_timing = off
+#track_wait_timing = off
 #track_functions = none			# none, pl, all
 #track_activity_query_size = 1024	# (change requires restart)
 #stats_temp_directory = 'pg_stat_tmp'
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index 427faa3..4e5a502 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5159,6 +5159,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,sslcompression,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '2228',
+  descr => 'statistics: information about accumulative data of wait event',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 's', proparallel => 'r',
+  prorettype => 'record', proargtypes => 'int4',
+  proallargtypes => '{int4,text,text,int8,int8}',
+  proargmodes => '{i,o,o,o,o}',
+  proargnames => '{pid,wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index e5a5d02..f90bb44 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -59,6 +59,7 @@ typedef enum StatMsgType
 	PGSTAT_MTYPE_ANALYZE,
 	PGSTAT_MTYPE_ARCHIVER,
 	PGSTAT_MTYPE_BGWRITER,
+	PGSTAT_MTYPE_WAITACCUM,
 	PGSTAT_MTYPE_FUNCSTAT,
 	PGSTAT_MTYPE_FUNCPURGE,
 	PGSTAT_MTYPE_RECOVERYCONFLICT,
@@ -119,7 +120,8 @@ typedef struct PgStat_TableCounts
 typedef enum PgStat_Shared_Reset_Target
 {
 	RESET_ARCHIVER,
-	RESET_BGWRITER
+	RESET_BGWRITER,
+	RESET_WAITACCUM
 } PgStat_Shared_Reset_Target;
 
 /* Possible object types for resetting single counters */
@@ -423,6 +425,33 @@ typedef struct PgStat_MsgBgWriter
 } PgStat_MsgBgWriter;
 
 /* ----------
+ * PgStat_WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct PgStat_WaitAccumEntry
+{
+	uint32			wait_event_info;
+	PgStat_Counter	calls;
+	instr_time		times;
+} PgStat_WaitAccumEntry;
+
+/* ----------
+ * PgStat_MsgWaitAccum	Sent by backend/background's process to update statistics.
+ * ----------
+ */
+#define PGSTAT_NUM_WAITACCUMENTRIES	\
+	((PGSTAT_MSG_PAYLOAD - sizeof(int))  \
+	 / sizeof(PgStat_WaitAccumEntry))
+
+typedef struct PgStat_MsgWaitAccum
+{
+	PgStat_MsgHdr m_hdr;
+
+	int m_nentries;
+	PgStat_WaitAccumEntry m_entry[PGSTAT_NUM_WAITACCUMENTRIES];
+} PgStat_MsgWaitAccum;
+
+/* ----------
  * PgStat_MsgRecoveryConflict	Sent by the backend upon recovery conflict
  * ----------
  */
@@ -564,6 +593,7 @@ typedef union PgStat_Msg
 	PgStat_MsgAnalyze msg_analyze;
 	PgStat_MsgArchiver msg_archiver;
 	PgStat_MsgBgWriter msg_bgwriter;
+	PgStat_MsgWaitAccum msg_waitaccum;
 	PgStat_MsgFuncstat msg_funcstat;
 	PgStat_MsgFuncpurge msg_funcpurge;
 	PgStat_MsgRecoveryConflict msg_recoveryconflict;
@@ -581,7 +611,7 @@ typedef union PgStat_Msg
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9D
+#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9E
 
 /* ----------
  * PgStat_StatDBEntry			The collector's data per database
@@ -711,6 +741,30 @@ typedef struct PgStat_GlobalStats
 	TimestampTz stat_reset_timestamp;
 } PgStat_GlobalStats;
 
+typedef struct WAEntry
+{
+	int key;
+	PgStat_WaitAccumEntry *entry;
+	struct WAEntry *next;
+} WAEntry;
+
+#define WA_BUCKET_SIZE 461
+
+typedef struct WAHash
+{
+	WAEntry entries[WA_BUCKET_SIZE];
+	WAEntry *buckets[WA_BUCKET_SIZE];
+	int entry_num;
+} WAHash;
+
+/*
+ * WaitAccum statistics kept in the stats collector
+ */
+typedef struct PgStat_WaitAccumStats
+{
+	WAHash *hash;
+} PgStat_WaitAccumStats;
+
 
 /* ----------
  * Backend types
@@ -787,6 +841,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+
 /* ----------
  * Wait Events - Client
  *
@@ -808,6 +864,8 @@ typedef enum
 	WAIT_EVENT_GSS_OPEN_SERVER,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_GSS_OPEN_SERVER
+
 /* ----------
  * Wait Events - IPC
  *
@@ -856,6 +914,8 @@ typedef enum
 	WAIT_EVENT_SYNC_REP
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_SYNC_REP
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -869,6 +929,8 @@ typedef enum
 	WAIT_EVENT_RECOVERY_APPLY_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_RECOVERY_APPLY_DELAY
+
 /* ----------
  * Wait Events - IO
  *
@@ -948,6 +1010,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_WAL_WRITE
+
 /* ----------
  * Command type for progress reporting purposes
  * ----------
@@ -1204,6 +1268,8 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
+extern WAHash *wa_hash;
+extern instr_time waitStart;
 
 /* ----------
  * GUC parameters
@@ -1211,6 +1277,7 @@ typedef struct PgStat_FunctionCallUsage
  */
 extern bool pgstat_track_activities;
 extern bool pgstat_track_counts;
+extern bool pgstat_track_wait_timing;
 extern int	pgstat_track_functions;
 extern PGDLLIMPORT int pgstat_track_activity_query_size;
 extern char *pgstat_stat_directory;
@@ -1228,6 +1295,7 @@ extern PgStat_MsgBgWriter BgWriterStats;
 extern PgStat_Counter pgStatBlockReadTime;
 extern PgStat_Counter pgStatBlockWriteTime;
 
+extern PgStat_WaitAccumEntry *pgstat_get_wa_entry(WAHash *hash, uint32 key);
 /* ----------
  * Functions called from postmaster
  * ----------
@@ -1315,6 +1383,50 @@ extern char *pgstat_clip_activity(const char *raw_activity);
  * initialized.
  * ----------
  */
+
+static inline void
+pgstat_report_waitaccum_start()
+{
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(waitStart);
+	}
+}
+
+static inline void
+pgstat_report_waitaccum_end(uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+	instr_time  diff;
+
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(diff);
+		INSTR_TIME_SUBTRACT(diff, waitStart);
+	}
+
+	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
+
+	if (!entry)
+	{
+		printf("wait_event_info: %u.\n", wait_event_info);
+		fflush(stdout);
+		return;
+	}
+
+	entry->calls++;
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+}
+
 static inline void
 pgstat_report_wait_start(uint32 wait_event_info)
 {
@@ -1328,6 +1440,8 @@ pgstat_report_wait_start(uint32 wait_event_info)
 	 * four-bytes, updates are atomic.
 	 */
 	proc->wait_event_info = wait_event_info;
+
+	pgstat_report_waitaccum_start();
 }
 
 /* ----------
@@ -1347,6 +1461,8 @@ pgstat_report_wait_end(void)
 	if (!pgstat_track_activities || !proc)
 		return;
 
+	pgstat_report_waitaccum_end(proc->wait_event_info);
+
 	/*
 	 * Since this is a four-byte field which is always read and written as
 	 * four-bytes, updates are atomic.
@@ -1354,6 +1470,7 @@ pgstat_report_wait_end(void)
 	proc->wait_event_info = 0;
 }
 
+
 /* nontransactional event counts are simple enough to inline */
 
 #define pgstat_count_heap_scan(rel)									\
@@ -1421,6 +1538,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 
 extern void pgstat_send_archiver(const char *xlog, bool failed);
 extern void pgstat_send_bgwriter(void);
+extern void pgstat_send_waitaccum(void);
 
 /* ----------
  * Support functions for the SQL-callable functions to
@@ -1435,5 +1553,6 @@ extern PgStat_StatFuncEntry *pgstat_fetch_stat_funcentry(Oid funcid);
 extern int	pgstat_fetch_stat_numbackends(void);
 extern PgStat_ArchiverStats *pgstat_fetch_stat_archiver(void);
 extern PgStat_GlobalStats *pgstat_fetch_global(void);
+extern PgStat_WaitAccumStats *pgstat_fetch_stat_waitaccum(void);
 
 #endif							/* PGSTAT_H */
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 8fda8e4..2149c96 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -184,6 +184,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 5b407e6..bd47ccb 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -21,6 +21,7 @@
 #include "storage/lock.h"
 #include "storage/pg_sema.h"
 #include "storage/proclist_types.h"
+#include "portability/instr_time.h"
 
 /*
  * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 62eaf90..82566d0 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2063,6 +2063,11 @@ pg_stat_user_tables| SELECT pg_stat_all_tables.relid,
     pg_stat_all_tables.autoanalyze_count
    FROM pg_stat_all_tables
   WHERE ((pg_stat_all_tables.schemaname <> ALL (ARRAY['pg_catalog'::name, 'information_schema'::name])) AND (pg_stat_all_tables.schemaname !~ '^pg_toast'::text));
+pg_stat_waitaccum| SELECT s.wait_event_type,
+    s.wait_event,
+    s.calls,
+    s.times
+   FROM pg_stat_get_waitaccum(NULL::integer) s(wait_event_type, wait_event, calls, times);
 pg_stat_wal_receiver| SELECT s.pid,
     s.status,
     s.receive_start_lsn,
-- 
1.8.3.1

From 4b9ac96fbf66222ae5fca60e5eed02209b42d1c8 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Wed, 15 Jan 2020 12:42:58 +0000
Subject: [PATCH v4 2/2] [POC] Change measuring method of wait event time from
 INSTR_TIME to rdtsc.

This patch changes measuring method of wait event time from INSTR_TIME (which
uses gettimeofday or clock_gettime) to rdtsc. This might reduce the overhead
of measuring overhead.

Any supports like changing clock cycle to actual time or error handling are
not currently implemented.
---
 src/include/pgstat.h | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index f90bb44..58fa1f7 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -432,7 +432,7 @@ typedef struct PgStat_WaitAccumEntry
 {
 	uint32			wait_event_info;
 	PgStat_Counter	calls;
-	instr_time		times;
+	uint64			times;
 } PgStat_WaitAccumEntry;
 
 /* ----------
@@ -1269,7 +1269,7 @@ typedef struct PgStat_FunctionCallUsage
 } PgStat_FunctionCallUsage;
 
 extern WAHash *wa_hash;
-extern instr_time waitStart;
+extern uint64 waitStart;
 
 /* ----------
  * GUC parameters
@@ -1392,7 +1392,7 @@ pgstat_report_waitaccum_start()
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(waitStart);
+		waitStart = rdtsc();
 	}
 }
 
@@ -1400,15 +1400,15 @@ static inline void
 pgstat_report_waitaccum_end(uint32 wait_event_info)
 {
 	PgStat_WaitAccumEntry *entry;
-	instr_time  diff;
+	uint64		diff = 0;
 
 	if (wa_hash == NULL)
 		return;
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(diff);
-		INSTR_TIME_SUBTRACT(diff, waitStart);
+		diff = rdtsc();
+		diff -= waitStart;
 	}
 
 	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
@@ -1423,7 +1423,7 @@ pgstat_report_waitaccum_end(uint32 wait_event_info)
 	entry->calls++;
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_ADD(entry->times, diff);
+		entry->times += diff;
 	}
 }
 
-- 
1.8.3.1

From 919ca45c748eff6f707233b62c735d8e173a1b86 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Wed, 15 Jan 2020 09:13:19 +0000
Subject: [PATCH v4 1/2] Add pg_stat_waitaccum view.

pg_stat_waitaccum shows counts and duration of each wait events.
Each backend/backgrounds counts and measures the time of wait event
in every pgstat_report_wait_start and pgstat_report_wait_end. They
store those info into their local variables and send to Statistics
Collector. We can get those info via Statistics Collector.

For reducing overhead, I implemented statistic hash instead of
dynamic hash. I also implemented track_wait_timing which
determines wait event duration is collected or not.

On windows, this function might be not worked correctly, because
now it initializes local variables in pg_stat_init which is not
passed to fork processes on windows.
---
 src/backend/catalog/system_views.sql          |   8 +
 src/backend/postmaster/pgstat.c               | 344 ++++++++++++++++++++++++++
 src/backend/storage/lmgr/lwlock.c             |  19 ++
 src/backend/utils/adt/pgstatfuncs.c           |  80 ++++++
 src/backend/utils/misc/guc.c                  |   9 +
 src/backend/utils/misc/postgresql.conf.sample |   1 +
 src/include/catalog/pg_proc.dat               |   9 +
 src/include/pgstat.h                          | 123 ++++++++-
 src/include/storage/lwlock.h                  |   1 +
 src/include/storage/proc.h                    |   1 +
 src/test/regress/expected/rules.out           |   5 +
 11 files changed, 598 insertions(+), 2 deletions(-)

diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index f681aaf..bc4faad 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -980,6 +980,14 @@ CREATE VIEW pg_stat_progress_analyze AS
     FROM pg_stat_get_progress_info('ANALYZE') AS S
         LEFT JOIN pg_database D ON S.datid = D.oid;
 
+CREATE VIEW pg_stat_waitaccum AS
+    SELECT
+		S.wait_event_type AS wait_event_type,
+		S.wait_event AS wait_event,
+		S.calls AS calls,
+		S.times AS times
+	FROM pg_stat_get_waitaccum(NULL) AS S;
+
 CREATE VIEW pg_stat_progress_vacuum AS
     SELECT
         S.pid AS pid, S.datid AS datid, D.datname AS datname,
diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 7169509..ceddfda 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -123,6 +123,7 @@
  */
 bool		pgstat_track_activities = false;
 bool		pgstat_track_counts = false;
+bool		pgstat_track_wait_timing = false;
 int			pgstat_track_functions = TRACK_FUNC_OFF;
 int			pgstat_track_activity_query_size = 1024;
 
@@ -153,6 +154,10 @@ static time_t last_pgstat_start_time;
 
 static bool pgStatRunningInCollector = false;
 
+WAHash *wa_hash;
+
+instr_time waitStart;
+
 /*
  * Structures in which backends store per-table info that's waiting to be
  * sent to the collector.
@@ -255,6 +260,7 @@ static int	localNumBackends = 0;
  */
 static PgStat_ArchiverStats archiverStats;
 static PgStat_GlobalStats globalStats;
+static PgStat_WaitAccumStats waitAccumStats;
 
 /*
  * List of OIDs of databases we need to write out.  If an entry is InvalidOid,
@@ -280,6 +286,8 @@ static pid_t pgstat_forkexec(void);
 #endif
 
 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
+static void pgstat_init_waitaccum_hash(WAHash **hash);
+static PgStat_WaitAccumEntry *pgstat_add_wa_entry(WAHash *hash, uint32 key);
 static void pgstat_beshutdown_hook(int code, Datum arg);
 
 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
@@ -287,8 +295,11 @@ static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
 												 Oid tableoid, bool create);
 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
+static void pgstat_write_waitaccum_statsfile(FILE *fpout);
 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
+static bool pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+											FILE *fpin, const char *statfile);
 static void backend_read_statsfile(void);
 static void pgstat_read_current_status(void);
 
@@ -324,6 +335,7 @@ static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
+static void pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len);
 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
@@ -331,6 +343,27 @@ static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
 static void pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len);
 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
 
+
+PgStat_WaitAccumEntry *
+pgstat_get_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *current;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	current = hash->buckets[bucket];
+
+	while (current != NULL)
+	{
+		if (current->key == key)
+			return current->entry;
+
+		current = current->next;
+	}
+
+	return NULL;
+}
+
+
 /* ------------------------------------------------------------
  * Public functions called from postmaster follow
  * ------------------------------------------------------------
@@ -602,6 +635,8 @@ retry2:
 
 	pg_freeaddrinfo_all(hints.ai_family, addrs);
 
+	pgstat_init_waitaccum_hash(&wa_hash);
+
 	return;
 
 startup_failed:
@@ -624,6 +659,75 @@ startup_failed:
 	SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
 }
 
+static PgStat_WaitAccumEntry *
+pgstat_add_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *prev;
+	WAEntry *new;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	prev = hash->buckets[bucket];
+
+	while (prev != NULL && prev->next != NULL)
+		prev = prev->next;
+
+	new = &hash->entries[hash->entry_num++];
+	new->key = key;
+	new->entry = MemoryContextAllocZero(TopMemoryContext, (sizeof(PgStat_WaitAccumEntry)));
+
+	if (prev != NULL)
+		prev->next = new;
+	else
+		hash->buckets[bucket] = new;
+
+	return new->entry;
+}
+
+static void
+pgstat_init_waitaccum_entry(WAHash *hash, uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+
+	entry = pgstat_add_wa_entry(hash, wait_event_info);
+	entry->wait_event_info = wait_event_info;
+}
+
+static void
+pgstat_init_waitaccum_hash(WAHash **hash)
+{
+	uint32 i;
+	int last_tranche_id;
+
+	*hash = MemoryContextAllocZero(TopMemoryContext, sizeof(WAHash));
+
+	last_tranche_id = LWLockGetLastTrancheId();
+	for (i = PG_WAIT_LWLOCK + 1; i <= (PG_WAIT_LWLOCK | last_tranche_id); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = (PG_WAIT_LOCK | LOCKTAG_RELATION); i <= (PG_WAIT_LOCK | LOCKTAG_LAST_TYPE); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_BUFFER_PIN; i <= PG_WAIT_BUFFER_PIN; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_ACTIVITY; i <= PG_WAIT_ACTIVITY_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_CLIENT; i <= PG_WAIT_CLIENT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	//do extension stuff
+
+	for (i = PG_WAIT_IPC; i <= PG_WAIT_IPC_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_TIMEOUT; i <= PG_WAIT_TIMEOUT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IO; i <= PG_WAIT_IO_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+}
+
 /*
  * subroutine for pgstat_reset_all
  */
@@ -904,6 +1008,9 @@ pgstat_report_stat(bool force)
 
 	/* Now, send function statistics */
 	pgstat_send_funcstats();
+
+	/* Send wait accumulative statistics */
+	pgstat_send_waitaccum();
 }
 
 /*
@@ -1334,6 +1441,8 @@ pgstat_reset_shared_counters(const char *target)
 		msg.m_resettarget = RESET_ARCHIVER;
 	else if (strcmp(target, "bgwriter") == 0)
 		msg.m_resettarget = RESET_BGWRITER;
+	else if (strcmp(target, "waitaccum") == 0)
+		msg.m_resettarget = RESET_WAITACCUM;
 	else
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
@@ -2618,6 +2727,22 @@ pgstat_fetch_global(void)
 	return &globalStats;
 }
 
+/*
+ * ---------
+ * pgstat_fetch_stat_waitaccum() -
+ *
+ *	Support function for the SQL-callable pgstat* functions. Returns
+ *	a pointer to the wait accum statistics struct.
+ * ---------
+ */
+PgStat_WaitAccumStats *
+pgstat_fetch_stat_waitaccum(void)
+{
+	backend_read_statsfile();
+
+	return &waitAccumStats;
+}
+
 
 /* ------------------------------------------------------------
  * Functions for management of the shared-memory PgBackendStatus array
@@ -4410,6 +4535,53 @@ pgstat_send_bgwriter(void)
 	MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
 }
 
+/* ----------
+ * pgstat_send_waitaccum() -
+ *
+ * ----------
+ */
+void
+pgstat_send_waitaccum()
+{
+	PgStat_MsgWaitAccum msg;
+	PgStat_WaitAccumEntry *entry;
+	int i;
+
+	if (wa_hash == NULL)
+		return;
+
+	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_WAITACCUM);
+	msg.m_nentries = 0;
+
+	for (i = 0; i < wa_hash->entry_num; i++)
+	{
+		entry = wa_hash->entries[i].entry;
+
+		/* Send only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Prepare and send the message
+		 */
+		memcpy(&msg.m_entry[msg.m_nentries], entry, sizeof(PgStat_WaitAccumEntry));
+		if (++msg.m_nentries >= PGSTAT_NUM_WAITACCUMENTRIES)
+		{
+			pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+						msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+			msg.m_nentries = 0;
+		}
+
+		/* Clear wait events information. */
+		entry->calls = 0;
+		INSTR_TIME_SET_ZERO(entry->times);
+	}
+
+	if (msg.m_nentries > 0)
+		pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+					msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+}
+
 
 /* ----------
  * PgstatCollectorMain() -
@@ -4600,6 +4772,10 @@ PgstatCollectorMain(int argc, char *argv[])
 					pgstat_recv_bgwriter(&msg.msg_bgwriter, len);
 					break;
 
+				case PGSTAT_MTYPE_WAITACCUM:
+					pgstat_recv_waitaccum(&msg.msg_waitaccum, len);
+					break;
+
 				case PGSTAT_MTYPE_FUNCSTAT:
 					pgstat_recv_funcstat(&msg.msg_funcstat, len);
 					break;
@@ -4868,6 +5044,8 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 	rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
 	(void) rc;					/* we'll check for error with ferror */
 
+	pgstat_write_waitaccum_statsfile(fpout);
+
 	/*
 	 * Walk through the database table.
 	 */
@@ -5073,6 +5251,43 @@ pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
 }
 
 /* ----------
+ * pgstat_write_waitaccum_statsfile() -
+ *		Write the waitAccumStats to the stat file.
+ *
+ * ----------
+ */
+static void
+pgstat_write_waitaccum_statsfile(FILE *fpout)
+{
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			rc;
+	int			i;
+
+	/*
+	 * Walk through the waitaccum hash.
+	 */
+	for (i = 0; i < hash->entry_num; i++)
+	{
+		entry = hash->entries[i].entry;
+
+		/* Write only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Write out the DB entry. We don't write the tables or functions
+		 * pointers, since they're of no use to any other process.
+		 */
+		fputc('D', fpout);
+		rc = fwrite(entry, sizeof(PgStat_WaitAccumEntry), 1, fpout);
+		(void) rc;				/* we'll check for error with ferror */
+	}
+
+	fputc('E', fpout);
+}
+
+/* ----------
  * pgstat_read_statsfiles() -
  *
  *	Reads in some existing statistics collector files and returns the
@@ -5125,6 +5340,7 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 	 */
 	memset(&globalStats, 0, sizeof(globalStats));
 	memset(&archiverStats, 0, sizeof(archiverStats));
+	waitAccumStats.hash = MemoryContextAllocZero(pgStatLocalContext, sizeof(WAHash));
 
 	/*
 	 * Set the current timestamp (will be kept only in case we can't load an
@@ -5195,6 +5411,9 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 		goto done;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&waitAccumStats, fpin, statfile))
+		goto done;
+
 	/*
 	 * We found an existing collector stats file. Read it and put all the
 	 * hashtable entries into place.
@@ -5493,10 +5712,13 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 	PgStat_StatDBEntry dbentry;
 	PgStat_GlobalStats myGlobalStats;
 	PgStat_ArchiverStats myArchiverStats;
+	PgStat_WaitAccumStats myWaitAccumStats;
 	FILE	   *fpin;
 	int32		format_id;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
 
+	myWaitAccumStats.hash = MemoryContextAllocZero(CurrentMemoryContext, sizeof(WAHash));
+
 	/*
 	 * Try to open the stats file.  As above, anything but ENOENT is worthy of
 	 * complaining about.
@@ -5547,6 +5769,9 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 		return false;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&myWaitAccumStats, fpin, statfile))
+		return false;
+
 	/* By default, we're going to return the timestamp of the global file. */
 	*ts = myGlobalStats.stats_timestamp;
 
@@ -5600,6 +5825,75 @@ done:
 	return true;
 }
 
+/* ----------
+ * pgstat_read_statsfiles() -
+ *
+ *	Reads the waitaccum stats from the file.
+ *	If an error happens when reading file, return false. Otherwise return true.
+ *
+ * ----------
+ */
+static bool
+pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+								FILE *fpin, const char *statfile)
+{
+	PgStat_WaitAccumEntry *entry;
+	PgStat_WaitAccumEntry buf;
+	WAHash *hash = stats->hash;
+
+	/*
+	 * Read and put all the hashtable entries into place.
+	 */
+	for (;;)
+	{
+		switch (fgetc(fpin))
+		{
+				/*
+				 * 'D'	A PgStat_WaitAccumEntry struct describing a database
+				 * follows.
+				 */
+			case 'D':
+				if (fread(&buf, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+									 != sizeof(PgStat_WaitAccumEntry))
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				entry = pgstat_get_wa_entry(hash, buf.wait_event_info);
+
+				if (entry)
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				/*
+				 * Add to the DB hash
+				 */
+				entry = pgstat_add_wa_entry(hash, buf.wait_event_info);
+				memcpy(entry, &buf, sizeof(PgStat_WaitAccumEntry));
+
+				break;
+
+			case 'E':
+				return true;
+
+			default:
+				ereport(pgStatRunningInCollector ? LOG : WARNING,
+						(errmsg("corrupted statistics file \"%s\"",
+								statfile)));
+				return false;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * If not already done, read the statistics collector stats file into
  * some hash tables.  The results will be kept until pgstat_clear_snapshot()
@@ -6109,7 +6403,20 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 		memset(&archiverStats, 0, sizeof(archiverStats));
 		archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
 	}
+	else if (msg->m_resettarget == RESET_WAITACCUM)
+	{
+		PgStat_WaitAccumEntry *entry;
+		WAHash *hash = waitAccumStats.hash;
+		int i;
+
+		for (i = 0; i < hash->entry_num; i++)
+		{
+			entry = hash->entries[i].entry;
 
+			entry->calls = 0;
+			INSTR_TIME_SET_ZERO(entry->times);
+		}
+	}
 	/*
 	 * Presumably the sender of this message validated the target, don't
 	 * complain here if it's not valid
@@ -6289,6 +6596,43 @@ pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
 }
 
 /* ----------
+ * pgstat_recv_waitaccum() -
+ *
+ *	Process a WAITACCUM message.
+ * ----------
+ */
+static void
+pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
+{
+	PgStat_WaitAccumEntry *m_entry = &(msg->m_entry[0]);
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			i;
+
+	/*
+	 * Process all function entries in the message.
+	 */
+	for (i = 0; i < msg->m_nentries; i++, m_entry++)
+	{
+		entry = pgstat_get_wa_entry(hash, m_entry->wait_event_info);
+
+		if (!entry)
+		{
+			entry = pgstat_add_wa_entry(hash, m_entry->wait_event_info);
+			memcpy(entry, m_entry, sizeof(PgStat_WaitAccumEntry));
+		}
+		else
+		{
+			/*
+			 * Otherwise add the values to the existing entry.
+			 */
+			entry->calls += m_entry->calls;
+			INSTR_TIME_ADD(entry->times, m_entry->times);
+		}
+	}
+}
+
+/* ----------
  * pgstat_recv_recoveryconflict() -
  *
  *	Process a RECOVERYCONFLICT message.
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 4c14e51..229993e 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -595,6 +595,25 @@ LWLockNewTrancheId(void)
 }
 
 /*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
+/*
  * Register a tranche ID in the lookup table for the current process.  This
  * routine will save a pointer to the tranche name passed as an argument,
  * so the name should be allocated in a backend-lifetime context
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 7e6a3c1..de0e1bf 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -1982,3 +1982,83 @@ pg_stat_get_archiver(PG_FUNCTION_ARGS)
 	/* Returns the record as Datum */
 	PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	PgStat_WaitAccumStats *waitaccum_stats;
+	PgStat_WaitAccumEntry *entry;
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	/* Get statistics about the waitaccum process */
+	waitaccum_stats = pgstat_fetch_stat_waitaccum();
+
+	for (i = 0; i < waitaccum_stats->hash->entry_num; i++)
+	{
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		entry = waitaccum_stats->hash->entries[i].entry;
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 8228e1f..8b07a97 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1483,6 +1483,15 @@ static struct config_bool ConfigureNamesBool[] =
 		false,
 		NULL, NULL, NULL
 	},
+	{
+		{"track_wait_timing", PGC_SUSET, STATS_COLLECTOR,
+			gettext_noop("Collects timing statistics for wait events."),
+			NULL
+		},
+		&pgstat_track_wait_timing,
+		false,
+		NULL, NULL, NULL
+	},
 
 	{
 		{"update_process_title", PGC_SUSET, PROCESS_TITLE,
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index e1048c0..3a99182 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -570,6 +570,7 @@
 #track_activities = on
 #track_counts = on
 #track_io_timing = off
+#track_wait_timing = off
 #track_functions = none			# none, pl, all
 #track_activity_query_size = 1024	# (change requires restart)
 #stats_temp_directory = 'pg_stat_tmp'
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index 226c904..a56baf3 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5179,6 +5179,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,sslcompression,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc,leader_pid}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '2228',
+  descr => 'statistics: information about accumulative data of wait event',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 's', proparallel => 'r',
+  prorettype => 'record', proargtypes => 'int4',
+  proallargtypes => '{int4,text,text,int8,int8}',
+  proargmodes => '{i,o,o,o,o}',
+  proargnames => '{pid,wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index aecb601..2f48282 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -59,6 +59,7 @@ typedef enum StatMsgType
 	PGSTAT_MTYPE_ANALYZE,
 	PGSTAT_MTYPE_ARCHIVER,
 	PGSTAT_MTYPE_BGWRITER,
+	PGSTAT_MTYPE_WAITACCUM,
 	PGSTAT_MTYPE_FUNCSTAT,
 	PGSTAT_MTYPE_FUNCPURGE,
 	PGSTAT_MTYPE_RECOVERYCONFLICT,
@@ -119,7 +120,8 @@ typedef struct PgStat_TableCounts
 typedef enum PgStat_Shared_Reset_Target
 {
 	RESET_ARCHIVER,
-	RESET_BGWRITER
+	RESET_BGWRITER,
+	RESET_WAITACCUM
 } PgStat_Shared_Reset_Target;
 
 /* Possible object types for resetting single counters */
@@ -423,6 +425,33 @@ typedef struct PgStat_MsgBgWriter
 } PgStat_MsgBgWriter;
 
 /* ----------
+ * PgStat_WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct PgStat_WaitAccumEntry
+{
+	uint32			wait_event_info;
+	PgStat_Counter	calls;
+	instr_time		times;
+} PgStat_WaitAccumEntry;
+
+/* ----------
+ * PgStat_MsgWaitAccum	Sent by backend/background's process to update statistics.
+ * ----------
+ */
+#define PGSTAT_NUM_WAITACCUMENTRIES	\
+	((PGSTAT_MSG_PAYLOAD - sizeof(int))  \
+	 / sizeof(PgStat_WaitAccumEntry))
+
+typedef struct PgStat_MsgWaitAccum
+{
+	PgStat_MsgHdr m_hdr;
+
+	int m_nentries;
+	PgStat_WaitAccumEntry m_entry[PGSTAT_NUM_WAITACCUMENTRIES];
+} PgStat_MsgWaitAccum;
+
+/* ----------
  * PgStat_MsgRecoveryConflict	Sent by the backend upon recovery conflict
  * ----------
  */
@@ -564,6 +593,7 @@ typedef union PgStat_Msg
 	PgStat_MsgAnalyze msg_analyze;
 	PgStat_MsgArchiver msg_archiver;
 	PgStat_MsgBgWriter msg_bgwriter;
+	PgStat_MsgWaitAccum msg_waitaccum;
 	PgStat_MsgFuncstat msg_funcstat;
 	PgStat_MsgFuncpurge msg_funcpurge;
 	PgStat_MsgRecoveryConflict msg_recoveryconflict;
@@ -581,7 +611,7 @@ typedef union PgStat_Msg
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9D
+#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9E
 
 /* ----------
  * PgStat_StatDBEntry			The collector's data per database
@@ -711,6 +741,30 @@ typedef struct PgStat_GlobalStats
 	TimestampTz stat_reset_timestamp;
 } PgStat_GlobalStats;
 
+typedef struct WAEntry
+{
+	int key;
+	PgStat_WaitAccumEntry *entry;
+	struct WAEntry *next;
+} WAEntry;
+
+#define WA_BUCKET_SIZE 461
+
+typedef struct WAHash
+{
+	WAEntry entries[WA_BUCKET_SIZE];
+	WAEntry *buckets[WA_BUCKET_SIZE];
+	int entry_num;
+} WAHash;
+
+/*
+ * WaitAccum statistics kept in the stats collector
+ */
+typedef struct PgStat_WaitAccumStats
+{
+	WAHash *hash;
+} PgStat_WaitAccumStats;
+
 
 /* ----------
  * Backend types
@@ -787,6 +841,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+
 /* ----------
  * Wait Events - Client
  *
@@ -808,6 +864,8 @@ typedef enum
 	WAIT_EVENT_GSS_OPEN_SERVER,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_GSS_OPEN_SERVER
+
 /* ----------
  * Wait Events - IPC
  *
@@ -856,6 +914,8 @@ typedef enum
 	WAIT_EVENT_SYNC_REP
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_SYNC_REP
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -869,6 +929,8 @@ typedef enum
 	WAIT_EVENT_RECOVERY_APPLY_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_RECOVERY_APPLY_DELAY
+
 /* ----------
  * Wait Events - IO
  *
@@ -948,6 +1010,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_WAL_WRITE
+
 /* ----------
  * Command type for progress reporting purposes
  * ----------
@@ -1205,6 +1269,8 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
+extern WAHash *wa_hash;
+extern instr_time waitStart;
 
 /* ----------
  * GUC parameters
@@ -1212,6 +1278,7 @@ typedef struct PgStat_FunctionCallUsage
  */
 extern PGDLLIMPORT bool pgstat_track_activities;
 extern PGDLLIMPORT bool pgstat_track_counts;
+extern PGDLLIMPORT bool pgstat_track_wait_timing;
 extern PGDLLIMPORT int pgstat_track_functions;
 extern PGDLLIMPORT int pgstat_track_activity_query_size;
 extern char *pgstat_stat_directory;
@@ -1229,6 +1296,7 @@ extern PgStat_MsgBgWriter BgWriterStats;
 extern PgStat_Counter pgStatBlockReadTime;
 extern PgStat_Counter pgStatBlockWriteTime;
 
+extern PgStat_WaitAccumEntry *pgstat_get_wa_entry(WAHash *hash, uint32 key);
 /* ----------
  * Functions called from postmaster
  * ----------
@@ -1316,6 +1384,50 @@ extern char *pgstat_clip_activity(const char *raw_activity);
  * initialized.
  * ----------
  */
+
+static inline void
+pgstat_report_waitaccum_start()
+{
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(waitStart);
+	}
+}
+
+static inline void
+pgstat_report_waitaccum_end(uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+	instr_time  diff;
+
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(diff);
+		INSTR_TIME_SUBTRACT(diff, waitStart);
+	}
+
+	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
+
+	if (!entry)
+	{
+		printf("wait_event_info: %u.\n", wait_event_info);
+		fflush(stdout);
+		return;
+	}
+
+	entry->calls++;
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+}
+
 static inline void
 pgstat_report_wait_start(uint32 wait_event_info)
 {
@@ -1329,6 +1441,8 @@ pgstat_report_wait_start(uint32 wait_event_info)
 	 * four-bytes, updates are atomic.
 	 */
 	proc->wait_event_info = wait_event_info;
+
+	pgstat_report_waitaccum_start();
 }
 
 /* ----------
@@ -1348,6 +1462,8 @@ pgstat_report_wait_end(void)
 	if (!pgstat_track_activities || !proc)
 		return;
 
+	pgstat_report_waitaccum_end(proc->wait_event_info);
+
 	/*
 	 * Since this is a four-byte field which is always read and written as
 	 * four-bytes, updates are atomic.
@@ -1355,6 +1471,7 @@ pgstat_report_wait_end(void)
 	proc->wait_event_info = 0;
 }
 
+
 /* nontransactional event counts are simple enough to inline */
 
 #define pgstat_count_heap_scan(rel)									\
@@ -1422,6 +1539,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 
 extern void pgstat_send_archiver(const char *xlog, bool failed);
 extern void pgstat_send_bgwriter(void);
+extern void pgstat_send_waitaccum(void);
 
 /* ----------
  * Support functions for the SQL-callable functions to
@@ -1436,5 +1554,6 @@ extern PgStat_StatFuncEntry *pgstat_fetch_stat_funcentry(Oid funcid);
 extern int	pgstat_fetch_stat_numbackends(void);
 extern PgStat_ArchiverStats *pgstat_fetch_stat_archiver(void);
 extern PgStat_GlobalStats *pgstat_fetch_global(void);
+extern PgStat_WaitAccumStats *pgstat_fetch_stat_waitaccum(void);
 
 #endif							/* PGSTAT_H */
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 8fda8e4..2149c96 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -184,6 +184,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index d217801..751f77a 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -21,6 +21,7 @@
 #include "storage/lock.h"
 #include "storage/pg_sema.h"
 #include "storage/proclist_types.h"
+#include "portability/instr_time.h"
 
 /*
  * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 634f825..2b98c4b 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2087,6 +2087,11 @@ pg_stat_user_tables| SELECT pg_stat_all_tables.relid,
     pg_stat_all_tables.autoanalyze_count
    FROM pg_stat_all_tables
   WHERE ((pg_stat_all_tables.schemaname <> ALL (ARRAY['pg_catalog'::name, 'information_schema'::name])) AND (pg_stat_all_tables.schemaname !~ '^pg_toast'::text));
+pg_stat_waitaccum| SELECT s.wait_event_type,
+    s.wait_event,
+    s.calls,
+    s.times
+   FROM pg_stat_get_waitaccum(NULL::integer) s(wait_event_type, wait_event, calls, times);
 pg_stat_wal_receiver| SELECT s.pid,
     s.status,
     s.receive_start_lsn,
-- 
1.8.3.1

From 4f89046235b95c5154a6ce38f6774573163e3ff0 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Wed, 15 Jan 2020 12:42:58 +0000
Subject: [PATCH v4 2/2] [POC] Change measuring method of wait event time from
 INSTR_TIME to rdtsc.

This patch changes measuring method of wait event time from INSTR_TIME (which
uses gettimeofday or clock_gettime) to rdtsc. This might reduce the overhead
of measuring overhead.

Any supports like changing clock cycle to actual time or error handling are
not currently implemented.
---
 src/include/pgstat.h | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 2f48282..4547bdb 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -432,7 +432,7 @@ typedef struct PgStat_WaitAccumEntry
 {
 	uint32			wait_event_info;
 	PgStat_Counter	calls;
-	instr_time		times;
+	uint64			times;
 } PgStat_WaitAccumEntry;
 
 /* ----------
@@ -1270,7 +1270,7 @@ typedef struct PgStat_FunctionCallUsage
 } PgStat_FunctionCallUsage;
 
 extern WAHash *wa_hash;
-extern instr_time waitStart;
+extern uint64 waitStart;
 
 /* ----------
  * GUC parameters
@@ -1393,7 +1393,7 @@ pgstat_report_waitaccum_start()
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(waitStart);
+		waitStart = rdtsc();
 	}
 }
 
@@ -1401,15 +1401,15 @@ static inline void
 pgstat_report_waitaccum_end(uint32 wait_event_info)
 {
 	PgStat_WaitAccumEntry *entry;
-	instr_time  diff;
+	uint64		diff = 0;
 
 	if (wa_hash == NULL)
 		return;
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(diff);
-		INSTR_TIME_SUBTRACT(diff, waitStart);
+		diff = rdtsc();
+		diff -= waitStart;
 	}
 
 	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
@@ -1424,7 +1424,7 @@ pgstat_report_waitaccum_end(uint32 wait_event_info)
 	entry->calls++;
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_ADD(entry->times, diff);
+		entry->times += diff;
 	}
 }
 
-- 
1.8.3.1

From 5bd0684dbb24dd5c76b8d35d1da33b247513a8b0 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Tue, 25 Feb 2020 07:24:01 +0000
Subject: [PATCH v6 1/2] Add pg_stat_waitaccum view.

pg_stat_waitaccum shows counts and duration of each wait events.
Each backend/backgrounds counts and measures the time of wait event
in every pgstat_report_wait_start and pgstat_report_wait_end. They
store those info into their local variables and send to Statistics
Collector. We can get those info via Statistics Collector.

For reducing overhead, I implemented statistic hash instead of
dynamic hash. I also implemented track_wait_timing which
determines wait event duration is collected or not.

On windows, this function might be not worked correctly, because
now it initializes local variables in pg_stat_init which is not
passed to fork processes on windows.
---
 src/backend/catalog/system_views.sql          |   8 +
 src/backend/postmaster/pgstat.c               | 344 ++++++++++++++++++++++++++
 src/backend/storage/lmgr/lwlock.c             |  19 ++
 src/backend/utils/adt/pgstatfuncs.c           |  80 ++++++
 src/backend/utils/misc/guc.c                  |   9 +
 src/backend/utils/misc/postgresql.conf.sample |   1 +
 src/include/catalog/pg_proc.dat               |   9 +
 src/include/pgstat.h                          | 123 ++++++++-
 src/include/storage/lwlock.h                  |   1 +
 src/include/storage/proc.h                    |   1 +
 src/test/regress/expected/rules.out           |   5 +
 11 files changed, 598 insertions(+), 2 deletions(-)

diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/system_views.sql
index f681aaf..bc4faad 100644
--- a/src/backend/catalog/system_views.sql
+++ b/src/backend/catalog/system_views.sql
@@ -980,6 +980,14 @@ CREATE VIEW pg_stat_progress_analyze AS
     FROM pg_stat_get_progress_info('ANALYZE') AS S
         LEFT JOIN pg_database D ON S.datid = D.oid;
 
+CREATE VIEW pg_stat_waitaccum AS
+    SELECT
+		S.wait_event_type AS wait_event_type,
+		S.wait_event AS wait_event,
+		S.calls AS calls,
+		S.times AS times
+	FROM pg_stat_get_waitaccum(NULL) AS S;
+
 CREATE VIEW pg_stat_progress_vacuum AS
     SELECT
         S.pid AS pid, S.datid AS datid, D.datname AS datname,
diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 59dc4f3..5c2f125 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -123,6 +123,7 @@
  */
 bool		pgstat_track_activities = false;
 bool		pgstat_track_counts = false;
+bool		pgstat_track_wait_timing = false;
 int			pgstat_track_functions = TRACK_FUNC_OFF;
 int			pgstat_track_activity_query_size = 1024;
 
@@ -153,6 +154,10 @@ static time_t last_pgstat_start_time;
 
 static bool pgStatRunningInCollector = false;
 
+WAHash *wa_hash;
+
+instr_time waitStart;
+
 /*
  * Structures in which backends store per-table info that's waiting to be
  * sent to the collector.
@@ -255,6 +260,7 @@ static int	localNumBackends = 0;
  */
 static PgStat_ArchiverStats archiverStats;
 static PgStat_GlobalStats globalStats;
+static PgStat_WaitAccumStats waitAccumStats;
 
 /*
  * List of OIDs of databases we need to write out.  If an entry is InvalidOid,
@@ -280,6 +286,8 @@ static pid_t pgstat_forkexec(void);
 #endif
 
 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
+static void pgstat_init_waitaccum_hash(WAHash **hash);
+static PgStat_WaitAccumEntry *pgstat_add_wa_entry(WAHash *hash, uint32 key);
 static void pgstat_beshutdown_hook(int code, Datum arg);
 
 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
@@ -287,8 +295,11 @@ static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
 												 Oid tableoid, bool create);
 static void pgstat_write_statsfiles(bool permanent, bool allDbs);
 static void pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent);
+static void pgstat_write_waitaccum_statsfile(FILE *fpout);
 static HTAB *pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep);
 static void pgstat_read_db_statsfile(Oid databaseid, HTAB *tabhash, HTAB *funchash, bool permanent);
+static bool pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+											FILE *fpin, const char *statfile);
 static void backend_read_statsfile(void);
 static void pgstat_read_current_status(void);
 
@@ -324,6 +335,7 @@ static void pgstat_recv_vacuum(PgStat_MsgVacuum *msg, int len);
 static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
+static void pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len);
 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
 static void pgstat_recv_funcpurge(PgStat_MsgFuncpurge *msg, int len);
 static void pgstat_recv_recoveryconflict(PgStat_MsgRecoveryConflict *msg, int len);
@@ -331,6 +343,27 @@ static void pgstat_recv_deadlock(PgStat_MsgDeadlock *msg, int len);
 static void pgstat_recv_checksum_failure(PgStat_MsgChecksumFailure *msg, int len);
 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
 
+
+PgStat_WaitAccumEntry *
+pgstat_get_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *current;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	current = hash->buckets[bucket];
+
+	while (current != NULL)
+	{
+		if (current->key == key)
+			return current->entry;
+
+		current = current->next;
+	}
+
+	return NULL;
+}
+
+
 /* ------------------------------------------------------------
  * Public functions called from postmaster follow
  * ------------------------------------------------------------
@@ -602,6 +635,8 @@ retry2:
 
 	pg_freeaddrinfo_all(hints.ai_family, addrs);
 
+	pgstat_init_waitaccum_hash(&wa_hash);
+
 	return;
 
 startup_failed:
@@ -624,6 +659,75 @@ startup_failed:
 	SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
 }
 
+static PgStat_WaitAccumEntry *
+pgstat_add_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *prev;
+	WAEntry *new;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	prev = hash->buckets[bucket];
+
+	while (prev != NULL && prev->next != NULL)
+		prev = prev->next;
+
+	new = &hash->entries[hash->entry_num++];
+	new->key = key;
+	new->entry = MemoryContextAllocZero(TopMemoryContext, (sizeof(PgStat_WaitAccumEntry)));
+
+	if (prev != NULL)
+		prev->next = new;
+	else
+		hash->buckets[bucket] = new;
+
+	return new->entry;
+}
+
+static void
+pgstat_init_waitaccum_entry(WAHash *hash, uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+
+	entry = pgstat_add_wa_entry(hash, wait_event_info);
+	entry->wait_event_info = wait_event_info;
+}
+
+static void
+pgstat_init_waitaccum_hash(WAHash **hash)
+{
+	uint32 i;
+	int last_tranche_id;
+
+	*hash = MemoryContextAllocZero(TopMemoryContext, sizeof(WAHash));
+
+	last_tranche_id = LWLockGetLastTrancheId();
+	for (i = PG_WAIT_LWLOCK + 1; i <= (PG_WAIT_LWLOCK | last_tranche_id); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = (PG_WAIT_LOCK | LOCKTAG_RELATION); i <= (PG_WAIT_LOCK | LOCKTAG_LAST_TYPE); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_BUFFER_PIN; i <= PG_WAIT_BUFFER_PIN; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_ACTIVITY; i <= PG_WAIT_ACTIVITY_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_CLIENT; i <= PG_WAIT_CLIENT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	//do extension stuff
+
+	for (i = PG_WAIT_IPC; i <= PG_WAIT_IPC_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_TIMEOUT; i <= PG_WAIT_TIMEOUT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IO; i <= PG_WAIT_IO_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+}
+
 /*
  * subroutine for pgstat_reset_all
  */
@@ -904,6 +1008,9 @@ pgstat_report_stat(bool force)
 
 	/* Now, send function statistics */
 	pgstat_send_funcstats();
+
+	/* Send wait accumulative statistics */
+	pgstat_send_waitaccum();
 }
 
 /*
@@ -1334,6 +1441,8 @@ pgstat_reset_shared_counters(const char *target)
 		msg.m_resettarget = RESET_ARCHIVER;
 	else if (strcmp(target, "bgwriter") == 0)
 		msg.m_resettarget = RESET_BGWRITER;
+	else if (strcmp(target, "waitaccum") == 0)
+		msg.m_resettarget = RESET_WAITACCUM;
 	else
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
@@ -2618,6 +2727,22 @@ pgstat_fetch_global(void)
 	return &globalStats;
 }
 
+/*
+ * ---------
+ * pgstat_fetch_stat_waitaccum() -
+ *
+ *	Support function for the SQL-callable pgstat* functions. Returns
+ *	a pointer to the wait accum statistics struct.
+ * ---------
+ */
+PgStat_WaitAccumStats *
+pgstat_fetch_stat_waitaccum(void)
+{
+	backend_read_statsfile();
+
+	return &waitAccumStats;
+}
+
 
 /* ------------------------------------------------------------
  * Functions for management of the shared-memory PgBackendStatus array
@@ -4410,6 +4535,53 @@ pgstat_send_bgwriter(void)
 	MemSet(&BgWriterStats, 0, sizeof(BgWriterStats));
 }
 
+/* ----------
+ * pgstat_send_waitaccum() -
+ *
+ * ----------
+ */
+void
+pgstat_send_waitaccum()
+{
+	PgStat_MsgWaitAccum msg;
+	PgStat_WaitAccumEntry *entry;
+	int i;
+
+	if (wa_hash == NULL)
+		return;
+
+	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_WAITACCUM);
+	msg.m_nentries = 0;
+
+	for (i = 0; i < wa_hash->entry_num; i++)
+	{
+		entry = wa_hash->entries[i].entry;
+
+		/* Send only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Prepare and send the message
+		 */
+		memcpy(&msg.m_entry[msg.m_nentries], entry, sizeof(PgStat_WaitAccumEntry));
+		if (++msg.m_nentries >= PGSTAT_NUM_WAITACCUMENTRIES)
+		{
+			pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+						msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+			msg.m_nentries = 0;
+		}
+
+		/* Clear wait events information. */
+		entry->calls = 0;
+		INSTR_TIME_SET_ZERO(entry->times);
+	}
+
+	if (msg.m_nentries > 0)
+		pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+					msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+}
+
 
 /* ----------
  * PgstatCollectorMain() -
@@ -4600,6 +4772,10 @@ PgstatCollectorMain(int argc, char *argv[])
 					pgstat_recv_bgwriter(&msg.msg_bgwriter, len);
 					break;
 
+				case PGSTAT_MTYPE_WAITACCUM:
+					pgstat_recv_waitaccum(&msg.msg_waitaccum, len);
+					break;
+
 				case PGSTAT_MTYPE_FUNCSTAT:
 					pgstat_recv_funcstat(&msg.msg_funcstat, len);
 					break;
@@ -4868,6 +5044,8 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 	rc = fwrite(&archiverStats, sizeof(archiverStats), 1, fpout);
 	(void) rc;					/* we'll check for error with ferror */
 
+	pgstat_write_waitaccum_statsfile(fpout);
+
 	/*
 	 * Walk through the database table.
 	 */
@@ -5073,6 +5251,43 @@ pgstat_write_db_statsfile(PgStat_StatDBEntry *dbentry, bool permanent)
 }
 
 /* ----------
+ * pgstat_write_waitaccum_statsfile() -
+ *		Write the waitAccumStats to the stat file.
+ *
+ * ----------
+ */
+static void
+pgstat_write_waitaccum_statsfile(FILE *fpout)
+{
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			rc;
+	int			i;
+
+	/*
+	 * Walk through the waitaccum hash.
+	 */
+	for (i = 0; i < hash->entry_num; i++)
+	{
+		entry = hash->entries[i].entry;
+
+		/* Write only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Write out the DB entry. We don't write the tables or functions
+		 * pointers, since they're of no use to any other process.
+		 */
+		fputc('D', fpout);
+		rc = fwrite(entry, sizeof(PgStat_WaitAccumEntry), 1, fpout);
+		(void) rc;				/* we'll check for error with ferror */
+	}
+
+	fputc('E', fpout);
+}
+
+/* ----------
  * pgstat_read_statsfiles() -
  *
  *	Reads in some existing statistics collector files and returns the
@@ -5125,6 +5340,7 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 	 */
 	memset(&globalStats, 0, sizeof(globalStats));
 	memset(&archiverStats, 0, sizeof(archiverStats));
+	waitAccumStats.hash = MemoryContextAllocZero(pgStatLocalContext, sizeof(WAHash));
 
 	/*
 	 * Set the current timestamp (will be kept only in case we can't load an
@@ -5195,6 +5411,9 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 		goto done;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&waitAccumStats, fpin, statfile))
+		goto done;
+
 	/*
 	 * We found an existing collector stats file. Read it and put all the
 	 * hashtable entries into place.
@@ -5493,10 +5712,13 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 	PgStat_StatDBEntry dbentry;
 	PgStat_GlobalStats myGlobalStats;
 	PgStat_ArchiverStats myArchiverStats;
+	PgStat_WaitAccumStats myWaitAccumStats;
 	FILE	   *fpin;
 	int32		format_id;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
 
+	myWaitAccumStats.hash = MemoryContextAllocZero(CurrentMemoryContext, sizeof(WAHash));
+
 	/*
 	 * Try to open the stats file.  As above, anything but ENOENT is worthy of
 	 * complaining about.
@@ -5547,6 +5769,9 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 		return false;
 	}
 
+	if(!pgstat_read_waitaccum_statsfile(&myWaitAccumStats, fpin, statfile))
+		return false;
+
 	/* By default, we're going to return the timestamp of the global file. */
 	*ts = myGlobalStats.stats_timestamp;
 
@@ -5600,6 +5825,75 @@ done:
 	return true;
 }
 
+/* ----------
+ * pgstat_read_statsfiles() -
+ *
+ *	Reads the waitaccum stats from the file.
+ *	If an error happens when reading file, return false. Otherwise return true.
+ *
+ * ----------
+ */
+static bool
+pgstat_read_waitaccum_statsfile(PgStat_WaitAccumStats *stats,
+								FILE *fpin, const char *statfile)
+{
+	PgStat_WaitAccumEntry *entry;
+	PgStat_WaitAccumEntry buf;
+	WAHash *hash = stats->hash;
+
+	/*
+	 * Read and put all the hashtable entries into place.
+	 */
+	for (;;)
+	{
+		switch (fgetc(fpin))
+		{
+				/*
+				 * 'D'	A PgStat_WaitAccumEntry struct describing a database
+				 * follows.
+				 */
+			case 'D':
+				if (fread(&buf, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+									 != sizeof(PgStat_WaitAccumEntry))
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				entry = pgstat_get_wa_entry(hash, buf.wait_event_info);
+
+				if (entry)
+				{
+					ereport(pgStatRunningInCollector ? LOG : WARNING,
+							(errmsg("corrupted statistics file \"%s\"",
+									statfile)));
+					return false;
+				}
+
+				/*
+				 * Add to the DB hash
+				 */
+				entry = pgstat_add_wa_entry(hash, buf.wait_event_info);
+				memcpy(entry, &buf, sizeof(PgStat_WaitAccumEntry));
+
+				break;
+
+			case 'E':
+				return true;
+
+			default:
+				ereport(pgStatRunningInCollector ? LOG : WARNING,
+						(errmsg("corrupted statistics file \"%s\"",
+								statfile)));
+				return false;
+		}
+	}
+
+	return 0;
+}
+
 /*
  * If not already done, read the statistics collector stats file into
  * some hash tables.  The results will be kept until pgstat_clear_snapshot()
@@ -6109,7 +6403,20 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 		memset(&archiverStats, 0, sizeof(archiverStats));
 		archiverStats.stat_reset_timestamp = GetCurrentTimestamp();
 	}
+	else if (msg->m_resettarget == RESET_WAITACCUM)
+	{
+		PgStat_WaitAccumEntry *entry;
+		WAHash *hash = waitAccumStats.hash;
+		int i;
+
+		for (i = 0; i < hash->entry_num; i++)
+		{
+			entry = hash->entries[i].entry;
 
+			entry->calls = 0;
+			INSTR_TIME_SET_ZERO(entry->times);
+		}
+	}
 	/*
 	 * Presumably the sender of this message validated the target, don't
 	 * complain here if it's not valid
@@ -6289,6 +6596,43 @@ pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len)
 }
 
 /* ----------
+ * pgstat_recv_waitaccum() -
+ *
+ *	Process a WAITACCUM message.
+ * ----------
+ */
+static void
+pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
+{
+	PgStat_WaitAccumEntry *m_entry = &(msg->m_entry[0]);
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			i;
+
+	/*
+	 * Process all function entries in the message.
+	 */
+	for (i = 0; i < msg->m_nentries; i++, m_entry++)
+	{
+		entry = pgstat_get_wa_entry(hash, m_entry->wait_event_info);
+
+		if (!entry)
+		{
+			entry = pgstat_add_wa_entry(hash, m_entry->wait_event_info);
+			memcpy(entry, m_entry, sizeof(PgStat_WaitAccumEntry));
+		}
+		else
+		{
+			/*
+			 * Otherwise add the values to the existing entry.
+			 */
+			entry->calls += m_entry->calls;
+			INSTR_TIME_ADD(entry->times, m_entry->times);
+		}
+	}
+}
+
+/* ----------
  * pgstat_recv_recoveryconflict() -
  *
  *	Process a RECOVERYCONFLICT message.
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 4c14e51..229993e 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -595,6 +595,25 @@ LWLockNewTrancheId(void)
 }
 
 /*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
+/*
  * Register a tranche ID in the lookup table for the current process.  This
  * routine will save a pointer to the tranche name passed as an argument,
  * so the name should be allocated in a backend-lifetime context
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 7e6a3c1..de0e1bf 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -1982,3 +1982,83 @@ pg_stat_get_archiver(PG_FUNCTION_ARGS)
 	/* Returns the record as Datum */
 	PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	PgStat_WaitAccumStats *waitaccum_stats;
+	PgStat_WaitAccumEntry *entry;
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	/* Get statistics about the waitaccum process */
+	waitaccum_stats = pgstat_fetch_stat_waitaccum();
+
+	for (i = 0; i < waitaccum_stats->hash->entry_num; i++)
+	{
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		entry = waitaccum_stats->hash->entries[i].entry;
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 8228e1f..8b07a97 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1483,6 +1483,15 @@ static struct config_bool ConfigureNamesBool[] =
 		false,
 		NULL, NULL, NULL
 	},
+	{
+		{"track_wait_timing", PGC_SUSET, STATS_COLLECTOR,
+			gettext_noop("Collects timing statistics for wait events."),
+			NULL
+		},
+		&pgstat_track_wait_timing,
+		false,
+		NULL, NULL, NULL
+	},
 
 	{
 		{"update_process_title", PGC_SUSET, PROCESS_TITLE,
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index e1048c0..3a99182 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -570,6 +570,7 @@
 #track_activities = on
 #track_counts = on
 #track_io_timing = off
+#track_wait_timing = off
 #track_functions = none			# none, pl, all
 #track_activity_query_size = 1024	# (change requires restart)
 #stats_temp_directory = 'pg_stat_tmp'
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index eb3c1a8..7b1b3d9 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5179,6 +5179,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,sslcompression,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc,leader_pid}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '2228',
+  descr => 'statistics: information about accumulative data of wait event',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 's', proparallel => 'r',
+  prorettype => 'record', proargtypes => 'int4',
+  proallargtypes => '{int4,text,text,int8,int8}',
+  proargmodes => '{i,o,o,o,o}',
+  proargnames => '{pid,wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 3a65a51..e5dbcb4 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -59,6 +59,7 @@ typedef enum StatMsgType
 	PGSTAT_MTYPE_ANALYZE,
 	PGSTAT_MTYPE_ARCHIVER,
 	PGSTAT_MTYPE_BGWRITER,
+	PGSTAT_MTYPE_WAITACCUM,
 	PGSTAT_MTYPE_FUNCSTAT,
 	PGSTAT_MTYPE_FUNCPURGE,
 	PGSTAT_MTYPE_RECOVERYCONFLICT,
@@ -119,7 +120,8 @@ typedef struct PgStat_TableCounts
 typedef enum PgStat_Shared_Reset_Target
 {
 	RESET_ARCHIVER,
-	RESET_BGWRITER
+	RESET_BGWRITER,
+	RESET_WAITACCUM
 } PgStat_Shared_Reset_Target;
 
 /* Possible object types for resetting single counters */
@@ -423,6 +425,33 @@ typedef struct PgStat_MsgBgWriter
 } PgStat_MsgBgWriter;
 
 /* ----------
+ * PgStat_WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct PgStat_WaitAccumEntry
+{
+	uint32			wait_event_info;
+	PgStat_Counter	calls;
+	instr_time		times;
+} PgStat_WaitAccumEntry;
+
+/* ----------
+ * PgStat_MsgWaitAccum	Sent by backend/background's process to update statistics.
+ * ----------
+ */
+#define PGSTAT_NUM_WAITACCUMENTRIES	\
+	((PGSTAT_MSG_PAYLOAD - sizeof(int))  \
+	 / sizeof(PgStat_WaitAccumEntry))
+
+typedef struct PgStat_MsgWaitAccum
+{
+	PgStat_MsgHdr m_hdr;
+
+	int m_nentries;
+	PgStat_WaitAccumEntry m_entry[PGSTAT_NUM_WAITACCUMENTRIES];
+} PgStat_MsgWaitAccum;
+
+/* ----------
  * PgStat_MsgRecoveryConflict	Sent by the backend upon recovery conflict
  * ----------
  */
@@ -564,6 +593,7 @@ typedef union PgStat_Msg
 	PgStat_MsgAnalyze msg_analyze;
 	PgStat_MsgArchiver msg_archiver;
 	PgStat_MsgBgWriter msg_bgwriter;
+	PgStat_MsgWaitAccum msg_waitaccum;
 	PgStat_MsgFuncstat msg_funcstat;
 	PgStat_MsgFuncpurge msg_funcpurge;
 	PgStat_MsgRecoveryConflict msg_recoveryconflict;
@@ -581,7 +611,7 @@ typedef union PgStat_Msg
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9D
+#define PGSTAT_FILE_FORMAT_ID	0x01A5BC9E
 
 /* ----------
  * PgStat_StatDBEntry			The collector's data per database
@@ -711,6 +741,30 @@ typedef struct PgStat_GlobalStats
 	TimestampTz stat_reset_timestamp;
 } PgStat_GlobalStats;
 
+typedef struct WAEntry
+{
+	int key;
+	PgStat_WaitAccumEntry *entry;
+	struct WAEntry *next;
+} WAEntry;
+
+#define WA_BUCKET_SIZE 461
+
+typedef struct WAHash
+{
+	WAEntry entries[WA_BUCKET_SIZE];
+	WAEntry *buckets[WA_BUCKET_SIZE];
+	int entry_num;
+} WAHash;
+
+/*
+ * WaitAccum statistics kept in the stats collector
+ */
+typedef struct PgStat_WaitAccumStats
+{
+	WAHash *hash;
+} PgStat_WaitAccumStats;
+
 
 /* ----------
  * Backend types
@@ -787,6 +841,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+
 /* ----------
  * Wait Events - Client
  *
@@ -808,6 +864,8 @@ typedef enum
 	WAIT_EVENT_WAL_SENDER_WRITE_DATA,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_GSS_OPEN_SERVER
+
 /* ----------
  * Wait Events - IPC
  *
@@ -856,6 +914,8 @@ typedef enum
 	WAIT_EVENT_SYNC_REP
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_SYNC_REP
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -869,6 +929,8 @@ typedef enum
 	WAIT_EVENT_RECOVERY_APPLY_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_RECOVERY_APPLY_DELAY
+
 /* ----------
  * Wait Events - IO
  *
@@ -948,6 +1010,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_WAL_WRITE
+
 /* ----------
  * Command type for progress reporting purposes
  * ----------
@@ -1205,6 +1269,8 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
+extern WAHash *wa_hash;
+extern instr_time waitStart;
 
 /* ----------
  * GUC parameters
@@ -1212,6 +1278,7 @@ typedef struct PgStat_FunctionCallUsage
  */
 extern PGDLLIMPORT bool pgstat_track_activities;
 extern PGDLLIMPORT bool pgstat_track_counts;
+extern PGDLLIMPORT bool pgstat_track_wait_timing;
 extern PGDLLIMPORT int pgstat_track_functions;
 extern PGDLLIMPORT int pgstat_track_activity_query_size;
 extern char *pgstat_stat_directory;
@@ -1229,6 +1296,7 @@ extern PgStat_MsgBgWriter BgWriterStats;
 extern PgStat_Counter pgStatBlockReadTime;
 extern PgStat_Counter pgStatBlockWriteTime;
 
+extern PgStat_WaitAccumEntry *pgstat_get_wa_entry(WAHash *hash, uint32 key);
 /* ----------
  * Functions called from postmaster
  * ----------
@@ -1316,6 +1384,50 @@ extern char *pgstat_clip_activity(const char *raw_activity);
  * initialized.
  * ----------
  */
+
+static inline void
+pgstat_report_waitaccum_start()
+{
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(waitStart);
+	}
+}
+
+static inline void
+pgstat_report_waitaccum_end(uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+	instr_time  diff;
+
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(diff);
+		INSTR_TIME_SUBTRACT(diff, waitStart);
+	}
+
+	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
+
+	if (!entry)
+	{
+		printf("wait_event_info: %u.\n", wait_event_info);
+		fflush(stdout);
+		return;
+	}
+
+	entry->calls++;
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+}
+
 static inline void
 pgstat_report_wait_start(uint32 wait_event_info)
 {
@@ -1329,6 +1441,8 @@ pgstat_report_wait_start(uint32 wait_event_info)
 	 * four-bytes, updates are atomic.
 	 */
 	proc->wait_event_info = wait_event_info;
+
+	pgstat_report_waitaccum_start();
 }
 
 /* ----------
@@ -1348,6 +1462,8 @@ pgstat_report_wait_end(void)
 	if (!pgstat_track_activities || !proc)
 		return;
 
+	pgstat_report_waitaccum_end(proc->wait_event_info);
+
 	/*
 	 * Since this is a four-byte field which is always read and written as
 	 * four-bytes, updates are atomic.
@@ -1355,6 +1471,7 @@ pgstat_report_wait_end(void)
 	proc->wait_event_info = 0;
 }
 
+
 /* nontransactional event counts are simple enough to inline */
 
 #define pgstat_count_heap_scan(rel)									\
@@ -1422,6 +1539,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 
 extern void pgstat_send_archiver(const char *xlog, bool failed);
 extern void pgstat_send_bgwriter(void);
+extern void pgstat_send_waitaccum(void);
 
 /* ----------
  * Support functions for the SQL-callable functions to
@@ -1436,5 +1554,6 @@ extern PgStat_StatFuncEntry *pgstat_fetch_stat_funcentry(Oid funcid);
 extern int	pgstat_fetch_stat_numbackends(void);
 extern PgStat_ArchiverStats *pgstat_fetch_stat_archiver(void);
 extern PgStat_GlobalStats *pgstat_fetch_global(void);
+extern PgStat_WaitAccumStats *pgstat_fetch_stat_waitaccum(void);
 
 #endif							/* PGSTAT_H */
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 8fda8e4..2149c96 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -184,6 +184,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index d217801..751f77a 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -21,6 +21,7 @@
 #include "storage/lock.h"
 #include "storage/pg_sema.h"
 #include "storage/proclist_types.h"
+#include "portability/instr_time.h"
 
 /*
  * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index 634f825..2b98c4b 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2087,6 +2087,11 @@ pg_stat_user_tables| SELECT pg_stat_all_tables.relid,
     pg_stat_all_tables.autoanalyze_count
    FROM pg_stat_all_tables
   WHERE ((pg_stat_all_tables.schemaname <> ALL (ARRAY['pg_catalog'::name, 'information_schema'::name])) AND (pg_stat_all_tables.schemaname !~ '^pg_toast'::text));
+pg_stat_waitaccum| SELECT s.wait_event_type,
+    s.wait_event,
+    s.calls,
+    s.times
+   FROM pg_stat_get_waitaccum(NULL::integer) s(wait_event_type, wait_event, calls, times);
 pg_stat_wal_receiver| SELECT s.pid,
     s.status,
     s.receive_start_lsn,
-- 
1.8.3.1

From cba515bd4d23e0b860f8b17a92932a3d7b3ab675 Mon Sep 17 00:00:00 2001
From: "imai.yoshikazu" <imai.yoshikazu@jp.fujitsu.com>
Date: Tue, 25 Feb 2020 07:25:26 +0000
Subject: [PATCH v6 2/2] [POC] Change measuring method of wait event time from
 INSTR_TIME to rdtsc.

This patch changes measuring method of wait event time from INSTR_TIME (which
uses gettimeofday or clock_gettime) to rdtsc. This might reduce the overhead
of measuring overhead.

Any supports like changing clock cycle to actual time or error handling are
not currently implemented.
---
 src/backend/postmaster/pgstat.c      |  8 ++++----
 src/backend/utils/adt/pgstatfuncs.c  |  2 +-
 src/include/pgstat.h                 | 14 +++++++-------
 src/include/portability/instr_time.h | 21 +++++++++++++++++++++
 4 files changed, 33 insertions(+), 12 deletions(-)

diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index 5c2f125..19c0017 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -156,7 +156,7 @@ static bool pgStatRunningInCollector = false;
 
 WAHash *wa_hash;
 
-instr_time waitStart;
+uint64 waitStart;
 
 /*
  * Structures in which backends store per-table info that's waiting to be
@@ -4574,7 +4574,7 @@ pgstat_send_waitaccum()
 
 		/* Clear wait events information. */
 		entry->calls = 0;
-		INSTR_TIME_SET_ZERO(entry->times);
+		entry->times = 0;
 	}
 
 	if (msg.m_nentries > 0)
@@ -6414,7 +6414,7 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 			entry = hash->entries[i].entry;
 
 			entry->calls = 0;
-			INSTR_TIME_SET_ZERO(entry->times);
+			entry->times = 0;
 		}
 	}
 	/*
@@ -6627,7 +6627,7 @@ pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
 			 * Otherwise add the values to the existing entry.
 			 */
 			entry->calls += m_entry->calls;
-			INSTR_TIME_ADD(entry->times, m_entry->times);
+			entry->times += m_entry->times;
 		}
 	}
 }
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index de0e1bf..9408c4b 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -2052,7 +2052,7 @@ pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
 
 		values[2] = Int64GetDatum(entry->calls);
 
-		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+		values[3] = UInt64GetDatum(entry->times);
 
 		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
 	}
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index e5dbcb4..962144d 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -432,7 +432,7 @@ typedef struct PgStat_WaitAccumEntry
 {
 	uint32			wait_event_info;
 	PgStat_Counter	calls;
-	instr_time		times;
+	uint64			times;
 } PgStat_WaitAccumEntry;
 
 /* ----------
@@ -1270,7 +1270,7 @@ typedef struct PgStat_FunctionCallUsage
 } PgStat_FunctionCallUsage;
 
 extern WAHash *wa_hash;
-extern instr_time waitStart;
+extern uint64 waitStart;
 
 /* ----------
  * GUC parameters
@@ -1393,7 +1393,7 @@ pgstat_report_waitaccum_start()
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(waitStart);
+		waitStart = rdtsc();
 	}
 }
 
@@ -1401,15 +1401,15 @@ static inline void
 pgstat_report_waitaccum_end(uint32 wait_event_info)
 {
 	PgStat_WaitAccumEntry *entry;
-	instr_time  diff;
+	uint64		diff = 0;
 
 	if (wa_hash == NULL)
 		return;
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(diff);
-		INSTR_TIME_SUBTRACT(diff, waitStart);
+		diff = rdtsc();
+		diff -= waitStart;
 	}
 
 	entry = pgstat_get_wa_entry(wa_hash, wait_event_info);
@@ -1424,7 +1424,7 @@ pgstat_report_waitaccum_end(uint32 wait_event_info)
 	entry->calls++;
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_ADD(entry->times, diff);
+		entry->times += diff;
 	}
 }
 
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index d645932..e3929c3 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -57,6 +57,10 @@
 
 #ifndef WIN32
 
+#if defined(__x86_64__) || defined(__i386__)
+#include <x86intrin.h>
+#endif
+
 #ifdef HAVE_CLOCK_GETTIME
 
 /* Use clock_gettime() */
@@ -209,6 +213,8 @@ typedef struct timeval instr_time;
 
 #else							/* WIN32 */
 
+#include <intrin.h>
+
 /* Use QueryPerformanceCounter() */
 
 typedef LARGE_INTEGER instr_time;
@@ -254,3 +260,18 @@ GetTimerFrequency(void)
 	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
 
 #endif							/* INSTR_TIME_H */
+
+#ifndef RDTSC_H_
+#define RDTSC_H_
+
+static inline uint64 rdtsc() {
+	uint64 result;
+#if defined(__x86_64__) || defined(__i386__) || defined(WIN32)
+	result = __rdtsc();
+#else
+	result = 0;
+#endif
+	return result;
+}
+
+#endif
-- 
1.8.3.1

From 88c2779679c5c9625ca5348eec0543daab5ccab4 Mon Sep 17 00:00:00 2001
From: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Date: Tue, 1 Jun 2021 13:25:57 +0200
Subject: [PATCH 1/2] Add pg_stat_waitaccum view.

pg_stat_waitaccum shows counts and duration of each wait events.
Each backend/backgrounds counts and measures the time of wait event
in every pgstat_report_wait_start and pgstat_report_wait_end. They
store those info into their local variables and send to Statistics
Collector. We can get those info via Statistics Collector.

For reducing overhead, I implemented statistic hash instead of
dynamic hash. I also implemented track_wait_timing which
determines wait event duration is collected or not.

On windows, this function might be not worked correctly, because
now it initializes local variables in pg_stat_init which is not
passed to fork processes on windows.
---
 src/backend/postmaster/pgstat.c               | 305 +++++++++++++++++-
 src/backend/storage/lmgr/lwlock.c             |  19 ++
 src/backend/utils/activity/wait_event.c       |  97 +++++-
 src/backend/utils/adt/pgstatfuncs.c           |  80 +++++
 src/backend/utils/misc/guc.c                  |  10 +
 src/backend/utils/misc/postgresql.conf.sample |   1 +
 src/include/catalog/pg_proc.dat               |   9 +
 src/include/pgstat.h                          |  68 +++-
 src/include/storage/lwlock.h                  |   1 +
 src/include/storage/proc.h                    |   1 +
 src/include/utils/wait_event.h                |  57 +---
 src/test/regress/expected/rules.out           |   5 +
 12 files changed, 599 insertions(+), 54 deletions(-)

diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index b0d07c0e0b..b4a84a2f62 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -114,6 +114,7 @@
  */
 bool		pgstat_track_counts = false;
 int			pgstat_track_functions = TRACK_FUNC_OFF;
+bool		pgstat_track_wait_timing = false;
 
 /* ----------
  * Built from GUC parameter
@@ -177,6 +178,10 @@ static time_t last_pgstat_start_time;
 
 static bool pgStatRunningInCollector = false;
 
+WAHash *wa_hash;
+
+instr_time waitStart;
+
 /*
  * Structures in which backends store per-table info that's waiting to be
  * sent to the collector.
@@ -276,6 +281,7 @@ static HTAB *pgStatDBHash = NULL;
  */
 static PgStat_ArchiverStats archiverStats;
 static PgStat_GlobalStats globalStats;
+static PgStat_WaitAccumStats waitAccumStats;
 static PgStat_WalStats walStats;
 static PgStat_SLRUStats slruStats[SLRU_NUM_ELEMENTS];
 static HTAB *replSlotStatHash = NULL;
@@ -305,6 +311,9 @@ static pid_t pgstat_forkexec(void);
 
 NON_EXEC_STATIC void PgstatCollectorMain(int argc, char *argv[]) pg_attribute_noreturn();
 
+static void pgstat_init_waitaccum_hash(WAHash **hash);
+static PgStat_WaitAccumEntry *pgstat_add_wa_entry(WAHash *hash, uint32 key);
+
 static PgStat_StatDBEntry *pgstat_get_db_entry(Oid databaseid, bool create);
 static PgStat_StatTabEntry *pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry,
 												 Oid tableoid, bool create);
@@ -348,6 +357,7 @@ static void pgstat_recv_analyze(PgStat_MsgAnalyze *msg, int len);
 static void pgstat_recv_anl_ancestors(PgStat_MsgAnlAncestors *msg, int len);
 static void pgstat_recv_archiver(PgStat_MsgArchiver *msg, int len);
 static void pgstat_recv_bgwriter(PgStat_MsgBgWriter *msg, int len);
+static void pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len);
 static void pgstat_recv_wal(PgStat_MsgWal *msg, int len);
 static void pgstat_recv_slru(PgStat_MsgSLRU *msg, int len);
 static void pgstat_recv_funcstat(PgStat_MsgFuncstat *msg, int len);
@@ -359,6 +369,25 @@ static void pgstat_recv_connstat(PgStat_MsgConn *msg, int len);
 static void pgstat_recv_replslot(PgStat_MsgReplSlot *msg, int len);
 static void pgstat_recv_tempfile(PgStat_MsgTempFile *msg, int len);
 
+PgStat_WaitAccumEntry *
+pgstat_get_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *current;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	current = hash->buckets[bucket];
+
+	while (current != NULL)
+	{
+		if (current->key == key)
+			return current->entry;
+
+		current = current->next;
+	}
+
+	return NULL;
+}
+
 /* ------------------------------------------------------------
  * Public functions called from postmaster follow
  * ------------------------------------------------------------
@@ -635,6 +664,8 @@ retry2:
 	/* Now that we have a long-lived socket, tell fd.c about it. */
 	ReserveExternalFD();
 
+	pgstat_init_waitaccum_hash(&wa_hash);
+
 	return;
 
 startup_failed:
@@ -657,6 +688,75 @@ startup_failed:
 	SetConfigOption("track_counts", "off", PGC_INTERNAL, PGC_S_OVERRIDE);
 }
 
+static PgStat_WaitAccumEntry *
+pgstat_add_wa_entry(WAHash *hash, uint32 key)
+{
+	WAEntry *prev;
+	WAEntry *new;
+	int bucket = key % WA_BUCKET_SIZE;
+
+	prev = hash->buckets[bucket];
+
+	while (prev != NULL && prev->next != NULL)
+		prev = prev->next;
+
+	new = &hash->entries[hash->entry_num++];
+	new->key = key;
+	new->entry = MemoryContextAllocZero(TopMemoryContext, (sizeof(PgStat_WaitAccumEntry)));
+
+	if (prev != NULL)
+		prev->next = new;
+	else
+		hash->buckets[bucket] = new;
+
+	return new->entry;
+}
+
+static void
+pgstat_init_waitaccum_entry(WAHash *hash, uint32 wait_event_info)
+{
+	PgStat_WaitAccumEntry *entry;
+
+	entry = pgstat_add_wa_entry(hash, wait_event_info);
+	entry->wait_event_info = wait_event_info;
+}
+
+static void
+pgstat_init_waitaccum_hash(WAHash **hash)
+{
+	uint32 i;
+	int last_tranche_id;
+
+	*hash = MemoryContextAllocZero(TopMemoryContext, sizeof(WAHash));
+
+	last_tranche_id = LWLockGetLastTrancheId();
+	for (i = PG_WAIT_LWLOCK + 1; i <= (PG_WAIT_LWLOCK | last_tranche_id); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = (PG_WAIT_LOCK | LOCKTAG_RELATION); i <= (PG_WAIT_LOCK | LOCKTAG_LAST_TYPE); i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_BUFFER_PIN; i <= PG_WAIT_BUFFER_PIN; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_ACTIVITY; i <= PG_WAIT_ACTIVITY_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_CLIENT; i <= PG_WAIT_CLIENT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IPC; i <= PG_WAIT_IPC_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_TIMEOUT; i <= PG_WAIT_TIMEOUT_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	for (i = PG_WAIT_IO; i <= PG_WAIT_IO_LAST_TYPE; i++)
+		pgstat_init_waitaccum_entry(*hash, i);
+
+	// FIXME: support extensions stuff
+}
+
 /*
  * subroutine for pgstat_reset_all
  */
@@ -960,8 +1060,11 @@ pgstat_report_stat(bool disconnect)
 	/* Send WAL statistics */
 	pgstat_send_wal(true);
 
-	/* Finally send SLRU statistics */
+	/* Send SLRU statistics */
 	pgstat_send_slru();
+
+	/* Finally send wait accumulative statistics */
+	pgstat_send_waitaccum();
 }
 
 /*
@@ -1453,6 +1556,8 @@ pgstat_reset_shared_counters(const char *target)
 		msg.m_resettarget = RESET_BGWRITER;
 	else if (strcmp(target, "wal") == 0)
 		msg.m_resettarget = RESET_WAL;
+	else if (strcmp(target, "waitaccum") == 0)
+		msg.m_resettarget = RESET_WAITACCUM;
 	else
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
@@ -2895,6 +3000,22 @@ pgstat_fetch_replslot(NameData slotname)
 	return pgstat_get_replslot_entry(slotname, false);
 }
 
+/*
+ * ---------
+ * pgstat_fetch_stat_waitaccum() -
+ *
+ *	Support function for the SQL-callable pgstat* functions. Returns
+ *	a pointer to the wait accum statistics struct.
+ * ---------
+ */
+PgStat_WaitAccumStats *
+pgstat_fetch_stat_waitaccum(void)
+{
+	backend_read_statsfile();
+
+	return &waitAccumStats;
+}
+
 /*
  * Shut down a single backend's statistics reporting at process exit.
  *
@@ -3172,6 +3293,52 @@ pgstat_send_slru(void)
 	}
 }
 
+/* ----------
+ * pgstat_send_waitaccum() -
+ *
+ * ----------
+ */
+void
+pgstat_send_waitaccum()
+{
+	PgStat_MsgWaitAccum msg;
+	PgStat_WaitAccumEntry *entry;
+	int i;
+
+	if (wa_hash == NULL)
+		return;
+
+	pgstat_setheader(&msg.m_hdr, PGSTAT_MTYPE_WAITACCUM);
+	msg.m_nentries = 0;
+
+	for (i = 0; i < wa_hash->entry_num; i++)
+	{
+		entry = wa_hash->entries[i].entry;
+
+		/* Send only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/*
+		 * Prepare and send the message
+		 */
+		memcpy(&msg.m_entry[msg.m_nentries], entry, sizeof(PgStat_WaitAccumEntry));
+		if (++msg.m_nentries >= PGSTAT_NUM_WAITACCUMENTRIES)
+		{
+			pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+						msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+			msg.m_nentries = 0;
+		}
+
+		/* Clear wait events information. */
+		entry->calls = 0;
+		INSTR_TIME_SET_ZERO(entry->times);
+	}
+
+	if (msg.m_nentries > 0)
+		pgstat_send(&msg, offsetof(PgStat_MsgWaitAccum, m_entry[0]) +
+					msg.m_nentries * sizeof(PgStat_WaitAccumEntry));
+}
 
 /* ----------
  * PgstatCollectorMain() -
@@ -3390,6 +3557,10 @@ PgstatCollectorMain(int argc, char *argv[])
 					pgstat_recv_slru(&msg.msg_slru, len);
 					break;
 
+				case PGSTAT_MTYPE_WAITACCUM:
+					pgstat_recv_waitaccum(&msg.msg_waitaccum, len);
+					break;
+
 				case PGSTAT_MTYPE_FUNCSTAT:
 					pgstat_recv_funcstat(&msg.msg_funcstat, len);
 					break;
@@ -3605,7 +3776,6 @@ pgstat_get_tab_entry(PgStat_StatDBEntry *dbentry, Oid tableoid, bool create)
 	return result;
 }
 
-
 /* ----------
  * pgstat_write_statsfiles() -
  *		Write the global statistics file, as well as requested DB files.
@@ -3629,7 +3799,7 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 	int32		format_id;
 	const char *tmpfile = permanent ? PGSTAT_STAT_PERMANENT_TMPFILE : pgstat_stat_tmpname;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
-	int			rc;
+	int			rc, i;
 
 	elog(DEBUG2, "writing stats file \"%s\"", statfile);
 
@@ -3725,6 +3895,23 @@ pgstat_write_statsfiles(bool permanent, bool allDbs)
 		}
 	}
 
+	/*
+	 * Walk through the waitaccum hash.
+	 */
+	for (i = 0; i < waitAccumStats.hash->entry_num; i++)
+	{
+		PgStat_WaitAccumEntry *entry = waitAccumStats.hash->entries[i].entry;
+
+		/* Write only wait events that have occurred. */
+		if (entry->calls == 0)
+			continue;
+
+		/* Write out the wait event entry */
+		fputc('W', fpout);
+		rc = fwrite(entry, sizeof(PgStat_WaitAccumEntry), 1, fpout);
+		(void) rc;				/* we'll check for error with ferror */
+	}
+
 	/*
 	 * No more output to be done. Close the temp file and replace the old
 	 * pgstat.stat with it.  The ferror() check replaces testing for error
@@ -3957,6 +4144,8 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 	memset(&archiverStats, 0, sizeof(archiverStats));
 	memset(&walStats, 0, sizeof(walStats));
 	memset(&slruStats, 0, sizeof(slruStats));
+	waitAccumStats.hash = MemoryContextAllocZero(pgStatLocalContext,
+												 sizeof(WAHash));
 
 	/*
 	 * Set the current timestamp (will be kept only in case we can't load an
@@ -4184,6 +4373,44 @@ pgstat_read_statsfiles(Oid onlydb, bool permanent, bool deep)
 					break;
 				}
 
+				/*
+				 * 'W'	A PgStat_WaitAccumEntry struct describing a Wait
+				 * event accumulator follows.
+				 */
+			case 'W':
+				{
+					PgStat_WaitAccumEntry *entry;
+					PgStat_WaitAccumEntry buf;
+					WAHash *hash = waitAccumStats.hash;
+
+					if (fread(&buf, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+										!= sizeof(PgStat_WaitAccumEntry))
+					{
+						ereport(pgStatRunningInCollector ? LOG : WARNING,
+								(errmsg("corrupted statistics file \"%s\"",
+										statfile)));
+						goto done;
+					}
+
+					entry = pgstat_get_wa_entry(hash, buf.wait_event_info);
+
+					if (entry)
+					{
+						ereport(pgStatRunningInCollector ? LOG : WARNING,
+								(errmsg("corrupted statistics file \"%s\"",
+										statfile)));
+						goto done;
+					}
+
+					/*
+					 * Add to the wait event hash
+					 */
+					entry = pgstat_add_wa_entry(hash, buf.wait_event_info);
+					memcpy(entry, &buf, sizeof(PgStat_WaitAccumEntry));
+
+					break;
+				}
+
 			case 'E':
 				goto done;
 
@@ -4396,6 +4623,7 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 	PgStat_WalStats myWalStats;
 	PgStat_SLRUStats mySLRUStats[SLRU_NUM_ELEMENTS];
 	PgStat_StatReplSlotEntry myReplSlotStats;
+	PgStat_WaitAccumEntry myWaitAccumStats;
 	FILE	   *fpin;
 	int32		format_id;
 	const char *statfile = permanent ? PGSTAT_STAT_PERMANENT_FILENAME : pgstat_stat_filename;
@@ -4526,6 +4754,26 @@ pgstat_read_db_statsfile_timestamp(Oid databaseid, bool permanent,
 				}
 				break;
 
+				/*
+				 * 'W'	A PgStat_WaitAccumEntry struct describing a Wait
+				 * event accumulator follows.
+				 */
+			case 'W':
+				{
+					if (fread(&myWaitAccumStats, 1, sizeof(PgStat_WaitAccumEntry), fpin)
+										!= sizeof(PgStat_WaitAccumEntry))
+					{
+						ereport(pgStatRunningInCollector ? LOG : WARNING,
+								(errmsg("corrupted statistics file \"%s\"",
+										statfile)));
+						FreeFile(fpin);
+						return false;
+					}
+
+					break;
+				}
+
+
 			case 'E':
 				goto done;
 
@@ -5071,6 +5319,20 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 		memset(&walStats, 0, sizeof(walStats));
 		walStats.stat_reset_timestamp = GetCurrentTimestamp();
 	}
+	else if (msg->m_resettarget == RESET_WAITACCUM)
+	{
+		PgStat_WaitAccumEntry *entry;
+		WAHash *hash = waitAccumStats.hash;
+		int i;
+
+		for (i = 0; i < hash->entry_num; i++)
+		{
+			entry = hash->entries[i].entry;
+
+			entry->calls = 0;
+			INSTR_TIME_SET_ZERO(entry->times);
+		}
+	}
 
 	/*
 	 * Presumably the sender of this message validated the target, don't
@@ -5394,6 +5656,43 @@ pgstat_recv_slru(PgStat_MsgSLRU *msg, int len)
 	slruStats[msg->m_index].truncate += msg->m_truncate;
 }
 
+/* ----------
+ * pgstat_recv_waitaccum() -
+ *
+ *	Process a WAITACCUM message.
+ * ----------
+ */
+static void
+pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
+{
+	PgStat_WaitAccumEntry *m_entry = &(msg->m_entry[0]);
+	PgStat_WaitAccumEntry *entry;
+	WAHash *hash = waitAccumStats.hash;
+	int			i;
+
+	/*
+	 * Process all function entries in the message.
+	 */
+	for (i = 0; i < msg->m_nentries; i++, m_entry++)
+	{
+		entry = pgstat_get_wa_entry(hash, m_entry->wait_event_info);
+
+		if (!entry)
+		{
+			entry = pgstat_add_wa_entry(hash, m_entry->wait_event_info);
+			memcpy(entry, m_entry, sizeof(PgStat_WaitAccumEntry));
+		}
+		else
+		{
+			/*
+			 * Otherwise add the values to the existing entry.
+			 */
+			entry->calls += m_entry->calls;
+			INSTR_TIME_ADD(entry->times, m_entry->times);
+		}
+	}
+}
+
 /* ----------
  * pgstat_recv_recoveryconflict() -
  *
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 55b9d7970e..e9a120805b 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -634,6 +634,25 @@ LWLockNewTrancheId(void)
 	return result;
 }
 
+/*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
 /*
  * Register a dynamic tranche name in the lookup table of the current process.
  *
diff --git a/src/backend/utils/activity/wait_event.c b/src/backend/utils/activity/wait_event.c
index 6baf67740c..392a783fae 100644
--- a/src/backend/utils/activity/wait_event.c
+++ b/src/backend/utils/activity/wait_event.c
@@ -25,6 +25,7 @@
 #include "storage/lmgr.h"		/* for GetLockNameFromTagType */
 #include "storage/lwlock.h"		/* for GetLWLockIdentifier */
 #include "utils/wait_event.h"
+#include "pgstat.h"
 
 
 static const char *pgstat_get_wait_activity(WaitEventActivity w);
@@ -32,7 +33,8 @@ static const char *pgstat_get_wait_client(WaitEventClient w);
 static const char *pgstat_get_wait_ipc(WaitEventIPC w);
 static const char *pgstat_get_wait_timeout(WaitEventTimeout w);
 static const char *pgstat_get_wait_io(WaitEventIO w);
-
+static inline void pgstat_report_waitaccum_start();
+static inline void pgstat_report_waitaccum_end();
 
 static uint32 local_my_wait_event_info;
 uint32	   *my_wait_event_info = &local_my_wait_event_info;
@@ -732,3 +734,96 @@ pgstat_get_wait_io(WaitEventIO w)
 
 	return event_name;
 }
+
+
+static inline void
+pgstat_report_waitaccum_start()
+{
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(waitStart);
+	}
+}
+
+static inline void
+pgstat_report_waitaccum_end()
+{
+	PgStat_WaitAccumEntry *entry;
+	instr_time  diff;
+
+	if (wa_hash == NULL)
+		return;
+
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_SET_CURRENT(diff);
+		INSTR_TIME_SUBTRACT(diff, waitStart);
+	}
+
+	entry = pgstat_get_wa_entry(wa_hash, *my_wait_event_info);
+
+	if (!entry)
+	{
+		return;
+	}
+
+	entry->calls++;
+	if (pgstat_track_wait_timing)
+	{
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+}
+
+
+/* ----------
+ * pgstat_report_wait_start() -
+ *
+ *	Called from places where server process needs to wait.  This is called
+ *	to report wait event information.  The wait information is stored
+ *	as 4-bytes where first byte represents the wait event class (type of
+ *	wait, for different types of wait, refer WaitClass) and the next
+ *	3-bytes represent the actual wait event.  Currently 2-bytes are used
+ *	for wait event which is sufficient for current usage, 1-byte is
+ *	reserved for future usage.
+ *
+ *	Historically we used to make this reporting conditional on
+ *	pgstat_track_activities, but the check for that seems to add more cost
+ *	than it saves.
+ *
+ *	my_wait_event_info initially points to local memory, making it safe to
+ *	call this before MyProc has been initialized.
+ * ----------
+ */
+inline void
+pgstat_report_wait_start(uint32 wait_event_info)
+{
+	/*
+	 * Since this is a four-byte field which is always read and written as
+	 * four-bytes, updates are atomic.
+	 */
+	*(volatile uint32 *) my_wait_event_info = wait_event_info;
+
+	//FIXME: recent patch to speed up this call.
+	pgstat_report_waitaccum_start();
+}
+
+/* ----------
+ * pgstat_report_wait_end() -
+ *
+ *	Called to report end of a wait.
+ * ----------
+ */
+inline void
+pgstat_report_wait_end(void)
+{
+	//FIXME: recent patch to speed up this call.
+	pgstat_report_waitaccum_end();
+
+	/* see pgstat_report_wait_start() */
+	*(volatile uint32 *) my_wait_event_info = 0;
+
+
+}
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 14056f5347..56afd20f22 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -2380,3 +2380,83 @@ pg_stat_get_replication_slot(PG_FUNCTION_ARGS)
 	/* Returns the record as Datum */
 	PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	PgStat_WaitAccumStats *waitaccum_stats;
+	PgStat_WaitAccumEntry *entry;
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	/* Get statistics about the waitaccum process */
+	waitaccum_stats = pgstat_fetch_stat_waitaccum();
+
+	for (i = 0; i < waitaccum_stats->hash->entry_num; i++)
+	{
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		entry = waitaccum_stats->hash->entries[i].entry;
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 68b62d523d..3260ab77d7 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -1546,6 +1546,16 @@ static struct config_bool ConfigureNamesBool[] =
 		NULL, NULL, NULL
 	},
 
+	{
+		{"track_wait_timing", PGC_SUSET, STATS_COLLECTOR,
+			gettext_noop("Collects timing statistics for wait events."),
+			NULL
+		},
+		&pgstat_track_wait_timing,
+		false,
+		NULL, NULL, NULL
+	},
+
 	{
 		{"update_process_title", PGC_SUSET, PROCESS_TITLE,
 			gettext_noop("Updates the process title to show the active SQL command."),
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index ddbb6dc2be..f6d0a5af8d 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -599,6 +599,7 @@
 #track_counts = on
 #track_io_timing = off
 #track_wal_io_timing = off
+#track_wait_timing = off
 #track_functions = none			# none, pl, all
 #stats_temp_directory = 'pg_stat_tmp'
 
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index acbcae4607..6c22bddc8e 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5284,6 +5284,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc,leader_pid,query_id}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '8316',
+  descr => 'statistics: information about accumulative data of wait event',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 's', proparallel => 'r',
+  prorettype => 'record', proargtypes => 'int4',
+  proallargtypes => '{int4,text,text,int8,int8}',
+  proargmodes => '{i,o,o,o,o}',
+  proargnames => '{pid,wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 9612c0a6c2..5564907df8 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -74,6 +74,7 @@ typedef enum StatMsgType
 	PGSTAT_MTYPE_BGWRITER,
 	PGSTAT_MTYPE_WAL,
 	PGSTAT_MTYPE_SLRU,
+	PGSTAT_MTYPE_WAITACCUM,
 	PGSTAT_MTYPE_FUNCSTAT,
 	PGSTAT_MTYPE_FUNCPURGE,
 	PGSTAT_MTYPE_RECOVERYCONFLICT,
@@ -137,14 +138,15 @@ typedef enum PgStat_Shared_Reset_Target
 {
 	RESET_ARCHIVER,
 	RESET_BGWRITER,
-	RESET_WAL
+	RESET_WAL,
+	RESET_WAITACCUM
 } PgStat_Shared_Reset_Target;
 
 /* Possible object types for resetting single counters */
 typedef enum PgStat_Single_Reset_Type
 {
 	RESET_TABLE,
-	RESET_FUNCTION
+	RESET_FUNCTION,
 } PgStat_Single_Reset_Type;
 
 /* ------------------------------------------------------------
@@ -539,6 +541,48 @@ typedef struct PgStat_MsgReplSlot
 	PgStat_Counter m_total_bytes;
 } PgStat_MsgReplSlot;
 
+/* ----------
+ * PgStat_WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct PgStat_WaitAccumEntry
+{
+	uint32			wait_event_info;
+	PgStat_Counter	calls;
+	instr_time		times;
+} PgStat_WaitAccumEntry;
+
+typedef struct WAEntry
+{
+	int key;
+	PgStat_WaitAccumEntry *entry;
+	struct WAEntry *next;
+} WAEntry;
+
+#define WA_BUCKET_SIZE 461
+
+typedef struct WAHash
+{
+	WAEntry entries[WA_BUCKET_SIZE];
+	WAEntry *buckets[WA_BUCKET_SIZE];
+	int entry_num;
+} WAHash;
+
+/* ----------
+ * PgStat_MsgWaitAccum	Sent by backend/background's process to update statistics.
+ * ----------
+ */
+#define PGSTAT_NUM_WAITACCUMENTRIES	\
+	((PGSTAT_MSG_PAYLOAD - sizeof(int))  \
+	 / sizeof(PgStat_WaitAccumEntry))
+
+typedef struct PgStat_MsgWaitAccum
+{
+	PgStat_MsgHdr m_hdr;
+
+	int m_nentries;
+	PgStat_WaitAccumEntry m_entry[PGSTAT_NUM_WAITACCUMENTRIES];
+} PgStat_MsgWaitAccum;
 
 /* ----------
  * PgStat_MsgRecoveryConflict	Sent by the backend upon recovery conflict
@@ -676,7 +720,6 @@ typedef struct PgStat_MsgConn
 	SessionEndType m_disconnect;
 } PgStat_MsgConn;
 
-
 /* ----------
  * PgStat_Msg					Union over all possible messages.
  * ----------
@@ -702,6 +745,7 @@ typedef union PgStat_Msg
 	PgStat_MsgBgWriter msg_bgwriter;
 	PgStat_MsgWal msg_wal;
 	PgStat_MsgSLRU msg_slru;
+	PgStat_MsgWaitAccum msg_waitaccum;
 	PgStat_MsgFuncstat msg_funcstat;
 	PgStat_MsgFuncpurge msg_funcpurge;
 	PgStat_MsgRecoveryConflict msg_recoveryconflict;
@@ -721,7 +765,7 @@ typedef union PgStat_Msg
  * ------------------------------------------------------------
  */
 
-#define PGSTAT_FILE_FORMAT_ID	0x01A5BCA2
+#define PGSTAT_FILE_FORMAT_ID	0x01A5BCA3
 
 /* ----------
  * PgStat_StatDBEntry			The collector's data per database
@@ -909,6 +953,15 @@ typedef struct PgStat_StatReplSlotEntry
 } PgStat_StatReplSlotEntry;
 
 
+/*
+ * WaitAccum statistics kept in the stats collector
+ */
+typedef struct PgStat_WaitAccumStats
+{
+	WAHash *hash;
+} PgStat_WaitAccumStats;
+
+
 /*
  * Working state needed to accumulate per-function-call timing statistics.
  */
@@ -925,6 +978,8 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
+extern instr_time waitStart;
+extern WAHash *wa_hash;
 
 /* ----------
  * GUC parameters
@@ -932,6 +987,7 @@ typedef struct PgStat_FunctionCallUsage
  */
 extern PGDLLIMPORT bool pgstat_track_counts;
 extern PGDLLIMPORT int pgstat_track_functions;
+extern PGDLLIMPORT bool pgstat_track_wait_timing;
 extern char *pgstat_stat_directory;
 extern char *pgstat_stat_tmpname;
 extern char *pgstat_stat_filename;
@@ -1092,6 +1148,7 @@ extern void pgstat_twophase_postabort(TransactionId xid, uint16 info,
 extern void pgstat_send_archiver(const char *xlog, bool failed);
 extern void pgstat_send_bgwriter(void);
 extern void pgstat_send_wal(bool force);
+extern void pgstat_send_waitaccum(void);
 
 /* ----------
  * Support functions for the SQL-callable functions to
@@ -1106,6 +1163,7 @@ extern PgStat_GlobalStats *pgstat_fetch_global(void);
 extern PgStat_WalStats *pgstat_fetch_stat_wal(void);
 extern PgStat_SLRUStats *pgstat_fetch_slru(void);
 extern PgStat_StatReplSlotEntry *pgstat_fetch_replslot(NameData slotname);
+extern PgStat_WaitAccumStats *pgstat_fetch_stat_waitaccum(void);
 
 extern void pgstat_count_slru_page_zeroed(int slru_idx);
 extern void pgstat_count_slru_page_hit(int slru_idx);
@@ -1117,4 +1175,6 @@ extern void pgstat_count_slru_truncate(int slru_idx);
 extern const char *pgstat_slru_name(int slru_idx);
 extern int	pgstat_slru_index(const char *name);
 
+extern PgStat_WaitAccumEntry *pgstat_get_wa_entry(WAHash *hash, uint32 key);
+
 #endif							/* PGSTAT_H */
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index a8f052e484..e6638b2c45 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -154,6 +154,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index be67d8a861..db09b3b64a 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -21,6 +21,7 @@
 #include "storage/lock.h"
 #include "storage/pg_sema.h"
 #include "storage/proclist_types.h"
+#include "portability/instr_time.h"
 
 /*
  * Each backend advertises up to PGPROC_MAX_CACHED_SUBXIDS TransactionIds
diff --git a/src/include/utils/wait_event.h b/src/include/utils/wait_event.h
index 6c6ec2e711..caa0db7f79 100644
--- a/src/include/utils/wait_event.h
+++ b/src/include/utils/wait_event.h
@@ -50,6 +50,8 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+
 /* ----------
  * Wait Events - Client
  *
@@ -70,6 +72,8 @@ typedef enum
 	WAIT_EVENT_WAL_SENDER_WRITE_DATA,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_WAL_SENDER_WRITE_DATA
+
 /* ----------
  * Wait Events - IPC
  *
@@ -128,6 +132,8 @@ typedef enum
 	WAIT_EVENT_XACT_GROUP_UPDATE
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_XACT_GROUP_UPDATE
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -143,6 +149,8 @@ typedef enum
 	WAIT_EVENT_VACUUM_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_VACUUM_DELAY
+
 /* ----------
  * Wait Events - IO
  *
@@ -227,58 +235,15 @@ typedef enum
 	WAIT_EVENT_LOGICAL_SUBXACT_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_LOGICAL_SUBXACT_WRITE
 
 extern const char *pgstat_get_wait_event(uint32 wait_event_info);
 extern const char *pgstat_get_wait_event_type(uint32 wait_event_info);
-static inline void pgstat_report_wait_start(uint32 wait_event_info);
-static inline void pgstat_report_wait_end(void);
+void pgstat_report_wait_start(uint32 wait_event_info);
+void pgstat_report_wait_end(void);
 extern void pgstat_set_wait_event_storage(uint32 *wait_event_info);
 extern void pgstat_reset_wait_event_storage(void);
 
 extern PGDLLIMPORT uint32 *my_wait_event_info;
 
-
-/* ----------
- * pgstat_report_wait_start() -
- *
- *	Called from places where server process needs to wait.  This is called
- *	to report wait event information.  The wait information is stored
- *	as 4-bytes where first byte represents the wait event class (type of
- *	wait, for different types of wait, refer WaitClass) and the next
- *	3-bytes represent the actual wait event.  Currently 2-bytes are used
- *	for wait event which is sufficient for current usage, 1-byte is
- *	reserved for future usage.
- *
- *	Historically we used to make this reporting conditional on
- *	pgstat_track_activities, but the check for that seems to add more cost
- *	than it saves.
- *
- *	my_wait_event_info initially points to local memory, making it safe to
- *	call this before MyProc has been initialized.
- * ----------
- */
-static inline void
-pgstat_report_wait_start(uint32 wait_event_info)
-{
-	/*
-	 * Since this is a four-byte field which is always read and written as
-	 * four-bytes, updates are atomic.
-	 */
-	*(volatile uint32 *) my_wait_event_info = wait_event_info;
-}
-
-/* ----------
- * pgstat_report_wait_end() -
- *
- *	Called to report end of a wait.
- * ----------
- */
-static inline void
-pgstat_report_wait_end(void)
-{
-	/* see pgstat_report_wait_start() */
-	*(volatile uint32 *) my_wait_event_info = 0;
-}
-
-
 #endif							/* WAIT_EVENT_H */
diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expected/rules.out
index e5ab11275d..4fbcefd46f 100644
--- a/src/test/regress/expected/rules.out
+++ b/src/test/regress/expected/rules.out
@@ -2183,6 +2183,11 @@ pg_stat_wal| SELECT w.wal_records,
     w.wal_sync_time,
     w.stats_reset
    FROM pg_stat_get_wal() w(wal_records, wal_fpi, wal_bytes, wal_buffers_full, wal_write, wal_sync, wal_write_time, wal_sync_time, stats_reset);
+pg_stat_waitaccum| SELECT s.wait_event_type,
+    s.wait_event,
+    s.calls,
+    s.times
+   FROM pg_stat_get_waitaccum(NULL::integer) s(wait_event_type, wait_event, calls, times);
 pg_stat_wal_receiver| SELECT s.pid,
     s.status,
     s.receive_start_lsn,
-- 
2.20.1

From ddb1adc5cd9acc9bc9de16d0cf057124b09fe1e3 Mon Sep 17 00:00:00 2001
From: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Date: Fri, 4 Jun 2021 18:14:51 +0200
Subject: [PATCH 2/2] [POC] Change measuring method of wait event time from
 INSTR_TIME to rdtsc.

This patch changes measuring method of wait event time from INSTR_TIME (which
uses gettimeofday or clock_gettime) to rdtsc. This might reduce the overhead
of measuring overhead.

Any supports like changing clock cycle to actual time or error handling are
not currently implemented.
---
 src/backend/postmaster/pgstat.c         |  8 ++++----
 src/backend/utils/activity/wait_event.c | 10 +++++-----
 src/backend/utils/adt/pgstatfuncs.c     |  2 +-
 src/include/pgstat.h                    |  4 ++--
 src/include/portability/instr_time.h    | 21 +++++++++++++++++++++
 5 files changed, 33 insertions(+), 12 deletions(-)

diff --git a/src/backend/postmaster/pgstat.c b/src/backend/postmaster/pgstat.c
index b4a84a2f62..e928239a29 100644
--- a/src/backend/postmaster/pgstat.c
+++ b/src/backend/postmaster/pgstat.c
@@ -180,7 +180,7 @@ static bool pgStatRunningInCollector = false;
 
 WAHash *wa_hash;
 
-instr_time waitStart;
+uint64 waitStart;
 
 /*
  * Structures in which backends store per-table info that's waiting to be
@@ -3332,7 +3332,7 @@ pgstat_send_waitaccum()
 
 		/* Clear wait events information. */
 		entry->calls = 0;
-		INSTR_TIME_SET_ZERO(entry->times);
+		entry->times = 0;
 	}
 
 	if (msg.m_nentries > 0)
@@ -5330,7 +5330,7 @@ pgstat_recv_resetsharedcounter(PgStat_MsgResetsharedcounter *msg, int len)
 			entry = hash->entries[i].entry;
 
 			entry->calls = 0;
-			INSTR_TIME_SET_ZERO(entry->times);
+			entry->times = 0;
 		}
 	}
 
@@ -5688,7 +5688,7 @@ pgstat_recv_waitaccum(PgStat_MsgWaitAccum *msg, int len)
 			 * Otherwise add the values to the existing entry.
 			 */
 			entry->calls += m_entry->calls;
-			INSTR_TIME_ADD(entry->times, m_entry->times);
+			entry->times += m_entry->times;
 		}
 	}
 }
diff --git a/src/backend/utils/activity/wait_event.c b/src/backend/utils/activity/wait_event.c
index 392a783fae..6d47eb0028 100644
--- a/src/backend/utils/activity/wait_event.c
+++ b/src/backend/utils/activity/wait_event.c
@@ -744,7 +744,7 @@ pgstat_report_waitaccum_start()
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(waitStart);
+		waitStart = rdtsc();
 	}
 }
 
@@ -752,15 +752,15 @@ static inline void
 pgstat_report_waitaccum_end()
 {
 	PgStat_WaitAccumEntry *entry;
-	instr_time  diff;
+	uint64		diff = 0;
 
 	if (wa_hash == NULL)
 		return;
 
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_SET_CURRENT(diff);
-		INSTR_TIME_SUBTRACT(diff, waitStart);
+		diff = rdtsc();
+		diff -= waitStart;
 	}
 
 	entry = pgstat_get_wa_entry(wa_hash, *my_wait_event_info);
@@ -773,7 +773,7 @@ pgstat_report_waitaccum_end()
 	entry->calls++;
 	if (pgstat_track_wait_timing)
 	{
-		INSTR_TIME_ADD(entry->times, diff);
+		entry->times += diff;
 	}
 }
 
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 56afd20f22..6faedd4938 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -2450,7 +2450,7 @@ pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
 
 		values[2] = Int64GetDatum(entry->calls);
 
-		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+		values[3] = UInt64GetDatum(entry->times);
 
 		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
 	}
diff --git a/src/include/pgstat.h b/src/include/pgstat.h
index 5564907df8..9a9ac8e16d 100644
--- a/src/include/pgstat.h
+++ b/src/include/pgstat.h
@@ -549,7 +549,7 @@ typedef struct PgStat_WaitAccumEntry
 {
 	uint32			wait_event_info;
 	PgStat_Counter	calls;
-	instr_time		times;
+	uint64			times;
 } PgStat_WaitAccumEntry;
 
 typedef struct WAEntry
@@ -978,7 +978,7 @@ typedef struct PgStat_FunctionCallUsage
 	instr_time	f_start;
 } PgStat_FunctionCallUsage;
 
-extern instr_time waitStart;
+extern uint64 waitStart;
 extern WAHash *wa_hash;
 
 /* ----------
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 39a4f0600e..507a1ca44d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -57,6 +57,10 @@
 
 #ifndef WIN32
 
+#if defined(__x86_64__) || defined(__i386__)
+#include <x86intrin.h>
+#endif
+
 #ifdef HAVE_CLOCK_GETTIME
 
 /* Use clock_gettime() */
@@ -209,6 +213,8 @@ typedef struct timeval instr_time;
 
 #else							/* WIN32 */
 
+#include <intrin.h>
+
 /* Use QueryPerformanceCounter() */
 
 typedef LARGE_INTEGER instr_time;
@@ -254,3 +260,18 @@ GetTimerFrequency(void)
 	(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
 
 #endif							/* INSTR_TIME_H */
+
+#ifndef RDTSC_H_
+#define RDTSC_H_
+
+static inline uint64 rdtsc() {
+	uint64 result;
+#if defined(__x86_64__) || defined(__i386__) || defined(WIN32)
+	result = __rdtsc();
+#else
+	result = 0;
+#endif
+	return result;
+}
+
+#endif
-- 
2.20.1

From e3d1f39138babfebd28a92f4dd8136eeb173d2c2 Mon Sep 17 00:00:00 2001
From: Jehan-Guillaume de Rorthais <jgdr@dalibo.com>
Date: Fri, 11 Jun 2021 18:17:47 +0200
Subject: [PATCH v1] PoC: Trace wait events to logfile when
 log_statement_stats=on

---
 src/backend/storage/lmgr/lwlock.c       |  19 +++
 src/backend/storage/lmgr/proc.c         |   2 +
 src/backend/tcop/postgres.c             |  48 ++++++-
 src/backend/utils/activity/wait_event.c | 158 ++++++++++++++++++++++++
 src/backend/utils/adt/pgstatfuncs.c     |  74 +++++++++++
 src/include/catalog/pg_proc.dat         |   9 ++
 src/include/storage/lock.h              |   1 +
 src/include/storage/lwlock.h            |   1 +
 src/include/utils/wait_event.h          |  77 ++++++++++++
 9 files changed, 387 insertions(+), 2 deletions(-)

diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 55b9d7970e..e9a120805b 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -634,6 +634,25 @@ LWLockNewTrancheId(void)
 	return result;
 }
 
+/*
+ * Get a last tranche ID.
+ */
+int
+LWLockGetLastTrancheId(void)
+{
+	int			result;
+	int		   *LWLockCounter;
+
+	Assert(!lock_named_request_allowed);
+
+	LWLockCounter = (int *) ((char *) MainLWLockArray - sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	result = *LWLockCounter;
+	SpinLockRelease(ShmemLock);
+
+	return result;
+}
+
 /*
  * Register a dynamic tranche name in the lookup table of the current process.
  *
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index 2575ea1ca0..52891c26a9 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -450,6 +450,8 @@ InitProcess(void)
 
 	/* now that we have a proc, report wait events to shared memory */
 	pgstat_set_wait_event_storage(&MyProc->wait_event_info);
+	/* init wait event tracking structure*/
+	pgstat_init_waitaccums();
 
 	/*
 	 * We might be reusing a semaphore that belonged to a failed process. So
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 8cea10c901..25500e8f76 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -207,7 +207,6 @@ static void log_disconnections(int code, Datum arg);
 static void enable_statement_timeout(void);
 static void disable_statement_timeout(void);
 
-
 /* ----------------------------------------------------------------
  *		routines to obtain user input
  * ----------------------------------------------------------------
@@ -978,7 +977,10 @@ exec_simple_query(const char *query_string)
 	 * results because ResetUsage wasn't called.
 	 */
 	if (save_log_statement_stats)
+	{
 		ResetUsage();
+		pgstat_set_report_waits();
+	}
 
 	/*
 	 * Start up a transaction command.  All queries generated by the
@@ -1308,7 +1310,10 @@ exec_simple_query(const char *query_string)
 	}
 
 	if (save_log_statement_stats)
+	{
 		ShowUsage("QUERY STATISTICS");
+		pgstat_reset_report_waits();
+	}
 
 	TRACE_POSTGRESQL_QUERY_DONE(query_string);
 
@@ -1346,7 +1351,10 @@ exec_parse_message(const char *query_string,	/* string to execute */
 	set_ps_display("PARSE");
 
 	if (save_log_statement_stats)
+	{
 		ResetUsage();
+		pgstat_set_report_waits();
+	}
 
 	ereport(DEBUG2,
 			(errmsg_internal("parse %s: %s",
@@ -1567,7 +1575,10 @@ exec_parse_message(const char *query_string,	/* string to execute */
 	}
 
 	if (save_log_statement_stats)
+	{
 		ShowUsage("PARSE MESSAGE STATISTICS");
+		pgstat_reset_report_waits();
+	}
 
 	debug_query_string = NULL;
 }
@@ -1637,7 +1648,10 @@ exec_bind_message(StringInfo input_message)
 	set_ps_display("BIND");
 
 	if (save_log_statement_stats)
+	{
 		ResetUsage();
+		pgstat_set_report_waits();
+	}
 
 	/*
 	 * Start up a transaction command so we can call functions etc. (Note that
@@ -2029,7 +2043,10 @@ exec_bind_message(StringInfo input_message)
 	}
 
 	if (save_log_statement_stats)
+	{
 		ShowUsage("BIND MESSAGE STATISTICS");
+		pgstat_reset_report_waits();
+	}
 
 	debug_query_string = NULL;
 }
@@ -2122,7 +2139,10 @@ exec_execute_message(const char *portal_name, long max_rows)
 	set_ps_display(GetCommandTagName(portal->commandTag));
 
 	if (save_log_statement_stats)
+	{
 		ResetUsage();
+		pgstat_set_report_waits();
+	}
 
 	BeginCommand(portal->commandTag, dest);
 
@@ -2269,7 +2289,10 @@ exec_execute_message(const char *portal_name, long max_rows)
 	}
 
 	if (save_log_statement_stats)
+	{
 		ShowUsage("EXECUTE MESSAGE STATISTICS");
+		pgstat_reset_report_waits();
+	}
 
 	debug_query_string = NULL;
 }
@@ -3914,7 +3937,6 @@ process_postgres_switches(int argc, char *argv[], GucContext ctx,
 #endif
 }
 
-
 /* ----------------------------------------------------------------
  * PostgresMain
  *	   postgres main loop -- all backends, interactive or otherwise start here
@@ -4795,10 +4817,13 @@ void
 ShowUsage(const char *title)
 {
 	StringInfoData str;
+	WaitAccumEntry *entry = wa_events.events;
 	struct timeval user,
 				sys;
 	struct timeval elapse_t;
 	struct rusage r;
+	bool		save_log_statement_stats = log_statement_stats;
+	int		i;
 
 	getrusage(RUSAGE_SELF, &r);
 	gettimeofday(&elapse_t, NULL);
@@ -4879,6 +4904,25 @@ ShowUsage(const char *title)
 					 r.ru_nvcsw, r.ru_nivcsw);
 #endif							/* HAVE_GETRUSAGE */
 
+	if (save_log_statement_stats)
+	{
+		for (i = 0; i < wa_events.num_events; i++, entry++)
+		{
+			const char *wait_event_type = NULL;
+			const char *wait_event = NULL;
+
+			if (entry->calls < 1)
+				continue;
+
+			wait_event_type = pgstat_get_wait_event_type(entry->wait_event_info);
+			wait_event = pgstat_get_wait_event(entry->wait_event_info);
+
+			appendStringInfo(&str, "!\t%s/%s %lu calls, %lu us elapsed\n",
+					 wait_event_type, wait_event, entry->calls,
+					 INSTR_TIME_GET_MICROSEC(entry->times));
+		}
+	}
+
 	/* remove trailing newline */
 	if (str.data[str.len - 1] == '\n')
 		str.data[--str.len] = '\0';
diff --git a/src/backend/utils/activity/wait_event.c b/src/backend/utils/activity/wait_event.c
index 6baf67740c..61dd125341 100644
--- a/src/backend/utils/activity/wait_event.c
+++ b/src/backend/utils/activity/wait_event.c
@@ -32,11 +32,127 @@ static const char *pgstat_get_wait_client(WaitEventClient w);
 static const char *pgstat_get_wait_ipc(WaitEventIPC w);
 static const char *pgstat_get_wait_timeout(WaitEventTimeout w);
 static const char *pgstat_get_wait_io(WaitEventIO w);
+static inline void accum_pgstat_report_wait_start();
+static inline void accum_pgstat_report_wait_end();
 
 
 static uint32 local_my_wait_event_info;
 uint32	   *my_wait_event_info = &local_my_wait_event_info;
 
+WaitEvents wa_events;
+instr_time waitEventStart;
+report_wait_start_function	my_pgstat_report_wait_start = standard_pgstat_report_wait_start;
+report_wait_end_function	my_pgstat_report_wait_end = standard_pgstat_report_wait_end;
+
+static WaitAccumEntry *
+pgstat_get_wait_entry(uint32 wait_event_info)
+{
+	uint32		classId;
+	uint16		eventId;
+	WaitAccumEntry *entry = NULL;
+
+	/* report process as not waiting. */
+	if (wait_event_info == 0)
+		return NULL;
+
+	classId = wait_event_info & 0xFF000000;
+	eventId = wait_event_info & 0x0000FFFF;
+
+	switch (classId)
+	{
+		case PG_WAIT_LWLOCK:
+			entry = wa_events.wa_lwlock + eventId;
+			break;
+		case PG_WAIT_LOCK:
+			entry = wa_events.wa_lock + eventId;
+			break;
+		case PG_WAIT_BUFFER_PIN:
+			entry = wa_events.wa_buffer_pin + eventId;
+			break;
+		case PG_WAIT_ACTIVITY:
+			entry = wa_events.wa_activity + eventId;
+			break;
+		case PG_WAIT_CLIENT:
+			entry = wa_events.wa_client + eventId;
+			break;
+		// FIXME
+		//  case PG_WAIT_EXTENSION:
+		//  	break;
+		case PG_WAIT_IPC:
+			entry = wa_events.wa_ipc + eventId;
+			break;
+		case PG_WAIT_TIMEOUT:
+			entry = wa_events.wa_timeout + eventId;
+			break;
+		case PG_WAIT_IO:
+			entry = wa_events.wa_io + eventId;
+			break;
+		default:
+			break;
+	}
+
+	return entry;
+}
+
+void
+pgstat_init_waitaccums(void)
+{
+	WaitAccumEntry *event;
+	int num_lwlocks;
+	int i;
+
+	num_lwlocks = LWLockGetLastTrancheId() +1;
+	wa_events.num_events = num_lwlocks
+		 + PG_WAIT_ACTIVITY_NUM
+		 + PG_WAIT_CLIENT_NUM
+		 + PG_WAIT_IPC_NUM
+		 + PG_WAIT_TIMEOUT_NUM
+		 + PG_WAIT_IO_NUM
+		 + LOCKTAG_NUM
+		 + PG_WAIT_BUFFER_PIN_NUM;
+
+	wa_events.events = MemoryContextAllocZero(TopMemoryContext,
+						  wa_events.num_events
+						  * sizeof(WaitAccumEntry));
+
+	event = wa_events.events;
+
+	// FIXME: support extensions
+
+	wa_events.wa_lwlock = event;
+	for (i = 0; i < num_lwlocks; i++, event++)
+		event->wait_event_info = PG_WAIT_LWLOCK|i;
+
+	wa_events.wa_lock = event;
+	for (i = 0; i < LOCKTAG_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_LOCK|i;
+
+	wa_events.wa_buffer_pin = event;
+	for (i = 0; i < PG_WAIT_BUFFER_PIN_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_BUFFER_PIN|i;
+
+	wa_events.wa_activity = event;
+	for (i = 0; i < PG_WAIT_ACTIVITY_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_ACTIVITY|i;
+
+	wa_events.wa_client = event;
+	for (i = 0; i < PG_WAIT_CLIENT_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_CLIENT|i;
+
+	wa_events.wa_ipc = event;
+	for (i = 0; i < PG_WAIT_IPC_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_IPC|i;
+
+	wa_events.wa_timeout = event;
+	for (i = 0; i < PG_WAIT_TIMEOUT_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_TIMEOUT|i;
+
+	wa_events.wa_io = event;
+	for (i = 0; i < PG_WAIT_IO_NUM; i++, event++)
+		event->wait_event_info = PG_WAIT_IO|i;
+
+	pgstat_reset_report_waits();
+}
 
 /*
  * Configure wait event reporting to report wait events to *wait_event_info.
@@ -732,3 +848,45 @@ pgstat_get_wait_io(WaitEventIO w)
 
 	return event_name;
 }
+
+static inline void
+accum_pgstat_report_wait_start(uint32 wait_event_info)
+{
+	standard_pgstat_report_wait_start(wait_event_info);
+
+	INSTR_TIME_SET_CURRENT(waitEventStart);
+}
+
+static inline void
+accum_pgstat_report_wait_end(void)
+{
+	WaitAccumEntry *entry;
+	instr_time  diff;
+
+	INSTR_TIME_SET_CURRENT(diff);
+	INSTR_TIME_SUBTRACT(diff, waitEventStart);
+
+	entry = pgstat_get_wait_entry(*my_wait_event_info);
+
+	if (entry)
+	{
+		entry->calls++;
+		INSTR_TIME_ADD(entry->times, diff);
+	}
+
+	standard_pgstat_report_wait_end();
+}
+
+void
+pgstat_set_report_waits(void)
+{
+	my_pgstat_report_wait_start = accum_pgstat_report_wait_start;
+	my_pgstat_report_wait_end = accum_pgstat_report_wait_end;
+}
+
+void
+pgstat_reset_report_waits(void)
+{
+	my_pgstat_report_wait_start = standard_pgstat_report_wait_start;
+	my_pgstat_report_wait_end = standard_pgstat_report_wait_end;
+}
diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pgstatfuncs.c
index 14056f5347..cf4db57bff 100644
--- a/src/backend/utils/adt/pgstatfuncs.c
+++ b/src/backend/utils/adt/pgstatfuncs.c
@@ -2380,3 +2380,77 @@ pg_stat_get_replication_slot(PG_FUNCTION_ARGS)
 	/* Returns the record as Datum */
 	PG_RETURN_DATUM(HeapTupleGetDatum(heap_form_tuple(tupdesc, values, nulls)));
 }
+
+Datum
+pg_stat_get_waitaccum(PG_FUNCTION_ARGS)
+{
+#define PG_STAT_GET_WAITACCUM_COLS	4
+	ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
+	TupleDesc	tupdesc;
+	Tuplestorestate *tupstore;
+	MemoryContext per_query_ctx;
+	MemoryContext oldcontext;
+	WaitAccumEntry *entry = wa_events.events;
+	int i;
+
+	/* check to see if caller supports us returning a tuplestore */
+	if (rsinfo == NULL || !IsA(rsinfo, ReturnSetInfo))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("set-valued function called in context that cannot accept a set")));
+	if (!(rsinfo->allowedModes & SFRM_Materialize))
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("materialize mode required, but it is not " \
+						"allowed in this context")));
+
+	/* Build a tuple descriptor for our result type */
+	if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+		elog(ERROR, "return type must be a row type");
+
+	per_query_ctx = rsinfo->econtext->ecxt_per_query_memory;
+	oldcontext = MemoryContextSwitchTo(per_query_ctx);
+
+	tupstore = tuplestore_begin_heap(true, false, work_mem);
+	rsinfo->returnMode = SFRM_Materialize;
+	rsinfo->setResult = tupstore;
+	rsinfo->setDesc = tupdesc;
+
+	MemoryContextSwitchTo(oldcontext);
+
+	for (i = 0; i < wa_events.num_events; i++, entry++)
+	{
+		Datum		values[PG_STAT_GET_WAITACCUM_COLS];
+		bool		nulls[PG_STAT_GET_WAITACCUM_COLS];
+		const char *wait_event_type = NULL;
+		const char *wait_event = NULL;
+
+		/* Initialise values and NULL flags arrays */
+		MemSet(values, 0, sizeof(values));
+		MemSet(nulls, 0, sizeof(nulls));
+
+		/* Fill values and NULLs */
+		{
+			uint32		raw_wait_event;
+
+			raw_wait_event = UINT32_ACCESS_ONCE(entry->wait_event_info);
+			wait_event_type = pgstat_get_wait_event_type(raw_wait_event);
+			wait_event = pgstat_get_wait_event(raw_wait_event);
+		}
+
+		values[0] = CStringGetTextDatum(wait_event_type);
+
+		values[1] = CStringGetTextDatum(wait_event);
+
+		values[2] = Int64GetDatum(entry->calls);
+
+		values[3] = UInt64GetDatum(INSTR_TIME_GET_MICROSEC(entry->times));
+
+		tuplestore_putvalues(tupstore, tupdesc, values, nulls);
+	}
+
+	/* clean up and return the tuplestore */
+	tuplestore_donestoring(tupstore);
+
+	return (Datum) 0;
+}
diff --git a/src/include/catalog/pg_proc.dat b/src/include/catalog/pg_proc.dat
index acbcae4607..b066863a97 100644
--- a/src/include/catalog/pg_proc.dat
+++ b/src/include/catalog/pg_proc.dat
@@ -5284,6 +5284,15 @@
   proargmodes => '{i,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o,o}',
   proargnames => '{pid,datid,pid,usesysid,application_name,state,query,wait_event_type,wait_event,xact_start,query_start,backend_start,state_change,client_addr,client_hostname,client_port,backend_xid,backend_xmin,backend_type,ssl,sslversion,sslcipher,sslbits,ssl_client_dn,ssl_client_serial,ssl_issuer_dn,gss_auth,gss_princ,gss_enc,leader_pid,query_id}',
   prosrc => 'pg_stat_get_activity' },
+{ oid => '8316',
+  descr => 'statistics: accumulated wait event stats',
+  proname => 'pg_stat_get_waitaccum', prorows => '200', proisstrict => 'f',
+  proretset => 't', provolatile => 'v', proparallel => 'r',
+  prorettype => 'record', proargtypes => '',
+  proallargtypes => '{text,text,int8,int8}',
+  proargmodes => '{o,o,o,o}',
+  proargnames => '{wait_event_type,wait_event,calls,times}',
+  prosrc => 'pg_stat_get_waitaccum' },
 { oid => '3318',
   descr => 'statistics: information about progress of backends running maintenance command',
   proname => 'pg_stat_get_progress_info', prorows => '100', proretset => 't',
diff --git a/src/include/storage/lock.h b/src/include/storage/lock.h
index 9b2a421c32..885664fa54 100644
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -152,6 +152,7 @@ typedef enum LockTagType
 } LockTagType;
 
 #define LOCKTAG_LAST_TYPE	LOCKTAG_ADVISORY
+#define	LOCKTAG_NUM 		LOCKTAG_LAST_TYPE
 
 extern const char *const LockTagTypeNames[];
 
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index a8f052e484..e6638b2c45 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -154,6 +154,7 @@ extern LWLockPadded *GetNamedLWLockTranche(const char *tranche_name);
  * registration in the main shared memory segment wouldn't work for that case.
  */
 extern int	LWLockNewTrancheId(void);
+extern int	LWLockGetLastTrancheId(void);
 extern void LWLockRegisterTranche(int tranche_id, const char *tranche_name);
 extern void LWLockInitialize(LWLock *lock, int tranche_id);
 
diff --git a/src/include/utils/wait_event.h b/src/include/utils/wait_event.h
index 6c6ec2e711..9da4b714e2 100644
--- a/src/include/utils/wait_event.h
+++ b/src/include/utils/wait_event.h
@@ -10,6 +10,8 @@
 #ifndef WAIT_EVENT_H
 #define WAIT_EVENT_H
 
+#include "utils/memutils.h"
+#include "portability/instr_time.h"
 
 /* ----------
  * Wait Classes
@@ -25,6 +27,40 @@
 #define PG_WAIT_TIMEOUT				0x09000000U
 #define PG_WAIT_IO					0x0A000000U
 
+/* ----------
+ * WaitAccumEntry	Entry in backend/background's per-wait_event_info hash table
+ * ----------
+ */
+typedef struct WaitAccumEntry
+{
+	uint32		wait_event_info;
+	uint64		calls;
+	instr_time	times;
+} WaitAccumEntry;
+
+typedef struct WaitEvents
+{
+	uint32		num_events;
+	WaitAccumEntry *events;
+	WaitAccumEntry *wa_lwlock;
+	WaitAccumEntry *wa_lock;
+	WaitAccumEntry *wa_buffer_pin;
+	WaitAccumEntry *wa_activity;
+	WaitAccumEntry *wa_client;
+	// FIXME WaitAccumEntry *wa_extension;
+	WaitAccumEntry *wa_ipc;
+	WaitAccumEntry *wa_timeout;
+	WaitAccumEntry *wa_io;
+} WaitEvents;
+
+extern WaitEvents wa_events;
+
+/*
+ * There's only one entry in PG_WAIT_BUFFER_PIN class:
+ * PG_WAIT_BUFFER_PIN itself
+ */
+#define PG_WAIT_BUFFER_PIN_NUM 1
+
 /* ----------
  * Wait Events - Activity
  *
@@ -50,6 +86,9 @@ typedef enum
 	WAIT_EVENT_WAL_WRITER_MAIN
 } WaitEventActivity;
 
+#define	PG_WAIT_ACTIVITY_LAST_TYPE	WAIT_EVENT_WAL_WRITER_MAIN
+#define	PG_WAIT_ACTIVITY_NUM ( WAIT_EVENT_WAL_WRITER_MAIN - PG_WAIT_ACTIVITY )
+
 /* ----------
  * Wait Events - Client
  *
@@ -70,6 +109,9 @@ typedef enum
 	WAIT_EVENT_WAL_SENDER_WRITE_DATA,
 } WaitEventClient;
 
+#define	PG_WAIT_CLIENT_LAST_TYPE	WAIT_EVENT_WAL_SENDER_WRITE_DATA
+#define	PG_WAIT_CLIENT_NUM ( PG_WAIT_CLIENT_LAST_TYPE - PG_WAIT_CLIENT )
+
 /* ----------
  * Wait Events - IPC
  *
@@ -128,6 +170,9 @@ typedef enum
 	WAIT_EVENT_XACT_GROUP_UPDATE
 } WaitEventIPC;
 
+#define	PG_WAIT_IPC_LAST_TYPE	WAIT_EVENT_XACT_GROUP_UPDATE
+#define	PG_WAIT_IPC_NUM ( PG_WAIT_IPC_LAST_TYPE - PG_WAIT_IPC )
+
 /* ----------
  * Wait Events - Timeout
  *
@@ -143,6 +188,9 @@ typedef enum
 	WAIT_EVENT_VACUUM_DELAY
 } WaitEventTimeout;
 
+#define	PG_WAIT_TIMEOUT_LAST_TYPE	WAIT_EVENT_VACUUM_DELAY
+#define	PG_WAIT_TIMEOUT_NUM ( PG_WAIT_TIMEOUT_LAST_TYPE - PG_WAIT_TIMEOUT )
+
 /* ----------
  * Wait Events - IO
  *
@@ -227,14 +275,30 @@ typedef enum
 	WAIT_EVENT_LOGICAL_SUBXACT_WRITE
 } WaitEventIO;
 
+#define	PG_WAIT_IO_LAST_TYPE	WAIT_EVENT_LOGICAL_SUBXACT_WRITE
+#define	PG_WAIT_IO_NUM ( PG_WAIT_IO_LAST_TYPE - PG_WAIT_IO )
+
+/* track start time of current wait event */
+extern instr_time waitEventStart;
+
+void pgstat_init_waitaccums(void);
 
 extern const char *pgstat_get_wait_event(uint32 wait_event_info);
 extern const char *pgstat_get_wait_event_type(uint32 wait_event_info);
 static inline void pgstat_report_wait_start(uint32 wait_event_info);
+static inline void standard_pgstat_report_wait_start(uint32 wait_event_info);
 static inline void pgstat_report_wait_end(void);
+static inline void standard_pgstat_report_wait_end(void);
+extern void pgstat_set_report_waits(void);
+extern void pgstat_reset_report_waits(void);
 extern void pgstat_set_wait_event_storage(uint32 *wait_event_info);
 extern void pgstat_reset_wait_event_storage(void);
 
+typedef void (*report_wait_start_function) (uint32 wait_event_info);
+extern PGDLLIMPORT report_wait_start_function my_pgstat_report_wait_start;
+typedef void (*report_wait_end_function) ();
+extern PGDLLIMPORT report_wait_end_function my_pgstat_report_wait_end;
+
 extern PGDLLIMPORT uint32 *my_wait_event_info;
 
 
@@ -259,6 +323,12 @@ extern PGDLLIMPORT uint32 *my_wait_event_info;
  */
 static inline void
 pgstat_report_wait_start(uint32 wait_event_info)
+{
+	(*my_pgstat_report_wait_start)(wait_event_info);
+}
+
+static inline void
+standard_pgstat_report_wait_start(uint32 wait_event_info)
 {
 	/*
 	 * Since this is a four-byte field which is always read and written as
@@ -275,6 +345,13 @@ pgstat_report_wait_start(uint32 wait_event_info)
  */
 static inline void
 pgstat_report_wait_end(void)
+{
+	(*my_pgstat_report_wait_end)();
+}
+
+
+static inline void
+standard_pgstat_report_wait_end(void)
 {
 	/* see pgstat_report_wait_start() */
 	*(volatile uint32 *) my_wait_event_info = 0;
-- 
2.20.1