model-writtenrun 7529 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
Dec 2018 · 2 entries
- Alexey Kondratov proposed changing tablespaces while CLUSTER, VACUUM FULL or REINDEX already rebuilds storage, allowing users to move the rebuilt relation onto another volume without a separate copy #1.
- Robert Haas proposed combining these operations with ALTER TABLE rewrites, while Alvaro Herrera explained the possible CLUSTER integration and the independent usefulness of REINDEX tablespace moves #2 #3 #6.
Aug - Sep 2019 · 2 entries
- After an eight-month gap, Alexey narrowed the implementation to REINDEX and REINDEX CONCURRENTLY, leaving the more complicated ALTER TABLE and CLUSTER integration unresolved #9.
- Surafel Temesgen requested NOWAIT, documentation and naming fixes; Alexey implemented them, then separated the optional SET TABLESPACE/NOWAIT syntax after Michael Paquier questioned coupling it to the main feature #10 #11 #12 #19.
Nov 2019 - Apr 2020 · 6 entries
- Steve Singer's testing exposed incorrect pg_default representation and incomplete NOWAIT locking behavior; Alexey dropped NOWAIT, added bulk REINDEX support and moved tablespace assignment into RelationSetNewRelfilenode() #20 #21.
- Masahiko Sawada and Michael reviewed mapped-relation checks, documentation, raw-node copying and completion; Michael then replaced the low-level relcache changes with explicit pg_class updates and command-counter advancement, which Alexey integrated #23 #24 #28 #30.
- Justin Pryzby restored the original CLUSTER/VACUUM FULL work and investigated crashes involving default tablespaces, repeated relfilenode disposal and system/TOAST permissions; Alexey consolidated these fixes into a cleaner series #34 #35 #37 #38.
- Justin proposed parenthesized options and an independent INDEX_TABLESPACE destination; after discussion of existing nonboolean options and extensibility, Michael supported parenthesized syntax and Justin implemented generic option lists #41 #44 #45 #47 #48.
- Alexey's review caught inconsistent option fields and unsafe validation changes, and he supplied a grammar-reuse delta; Justin incorporated it, removed the raw-name experiments and fixed boolean options such as VERBOSE OFF #51 #52 #53 #54.
- The feature missed the PostgreSQL 13 window and moved to the next development cycle, with restored REINDEX DATABASE tests exposing further warning and TOAST-handling fixes #56 #57.
Aug 2020 - Feb 2021 · 20 entries
- Michael resumed with the independent psql completion component, revising Justin's patch and removing an unwanted opening-parenthesis suggestion before landing 1f75b45413 from #58 #59 #60.
- The remaining series separated parsed options from execution flags, corrected relfilenode bookkeeping and disallowed mixing legacy CLUSTER VERBOSE syntax with parenthesized options; Alexey also discovered duplicate VACUUM and EXPLAIN option grammar #62 #71 #72.
- Partitioned-index support brought additional tablespace paths and tests, followed by fixes for repeated name lookup, default-tablespace normalization and no-op moves, plus a reusable tablespace-update helper #75 #81.
- After October and November rebases, Michael landed the shared EXPLAIN/VACUUM/ANALYZE grammar component from #83 as 873ea9ee69, renaming its common_option_* rules to utility_option_* #82 #84.
- Michael next isolated and polished CLUSTER/REINDEX DefElem parsing, relocating execution flags and the REINDEX parser helper and correcting syntax, documentation and tests; #86 landed unchanged as b5913f6120 #89.
- Justin introduced ExecReindex and a parameter structure to simplify the remaining feature; Alexey reviewed its bit manipulation, and Justin retained intermediate booleans because repeated options such as VERBOSE OFF must clear earlier settings #90 #95 #97.
- Peter Eisentraut objected to enums used as bitmasks, prompting Michael's utility cleanup and Peter's separate pg_dump macro conversion, which landed directly from discussion as d2a2808eb4 #92 #98 #99.
- Peter proposed boolean option structures instead; Justin incorporated them, repaired initialization and pointer-passing issues and extended the design to CLUSTER and VACUUM, while Alexey identified inconsistencies in the VACUUM helper APIs #102 #103 #106 #107 #108.
- Alvaro's ABI concerns redirected the design to extensible parameter structures containing hexadecimal bits32 masks; Justin then removed unnecessary allocations and memory-context handling in favor of stack-owned structures and assignment #109 #111 #112 #114 #117.
- During this refactoring, Zhihong Yu identified a redundant default-tablespace condition in the REINDEX consumer, and Justin and Alexey accepted the correction #115 #116 #118.
- The option API settled on shared parameter structures while retaining relation-local REINDEX flags separately; after Alexey's naming and type review, Michael landed #128 with light edits as a3dc926009 #119 #124 #130.
- Justin resent the four remaining tablespace patches, while Michael asked that the broader enum cleanup be pursued separately; Michael and Alvaro then requested ACL coverage, nested-partition tests and reuse of existing tablecmds.c tablespace logic #131 #132 #134 #136.
- Alexey moved the helper into tablecmds.c and added nested-partition tests that exposed a default-tablespace bug; discussion settled on leaving TOAST indexes in their existing tablespace when rebuilding their parent table's indexes #138 #142 #143 #144.
- Michael redesigned the helper into separate validation and catalog-update routines shared by relations with and without storage, also correcting tablespace dependency handling #148.
- The helper component of #149 landed as 4c9c359d38, with ALTER hooks retained in ALTER callers rather than the reusable routines #150.
- Alexey resent REINDEX with ACL rechecks and relfilenode tests, but Alvaro showed that ShareLock allowed concurrent metadata modifications; switching to AccessExclusiveLock then raised deadlock-ordering and REINDEX CONCURRENTLY concerns #151 #153 #155 #156.
- Alexey separated partitioned-index metadata updates from the core feature; Michael chose to move leaf indexes while leaving storage-less partitioned relations unchanged, documenting the additional ALTER command needed to update their tablespace metadata #157 #160.
- Michael landed c5b286047c from #160 after further system/TOAST restrictions, repair of an ALTER TABLE TOAST-cascade regression and additional tests #165.
- The REINDEX landing closed the commitfest entry, while CLUSTER, VACUUM FULL and their INDEX_TABLESPACE work remained uncommitted; Michael recommended separate threads and reconsideration of integration with ALTER TABLE's combined rewrites #165.
- Justin subsequently proposed removing apparently dead concurrent-REINDEX checks, but Michael demonstrated that direct TOAST table and index moves reach those checks, so the cleanup was rejected #166 #167.
Apr 2021 · 3 entries
- An independent late branch began when Alexey reported forced-TCP PostgresNode tests looping through ports on BSD/macOS and supplied a fix and reproducer #168.
- Tom Lane rejected unnecessary Linux loopback probes, Andrew Dunstan proposed testing extra addresses only when both Windows and TCP apply, and Alexey revised the condition and comments #169 #170 #171 #172.
- Andrew landed the PostgresNode fix from #172 with slight edits as d064afc720 and confirmed the push in the thread's final message; the earlier CLUSTER/VACUUM FULL extensions remained deferred #173 #165.