Dec 2021

1 commit
pg_dump: avoid unsafe function calls in getPolicies().
31 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.7REL_13_STABLE13.10+2 moreAlso in: [BUG] pg_dump blocked (written from the discussion, no matching patch)Also in: Assorted improvements in pg_dump (not this thread)

Dec 2021

6 commits
Account for TOAST data while scheduling parallel dumps.
6 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #141 of 1 reviewer saw the final version

The TOAST scheduling patch in #14 matches this commit's stable patch ID and normalized diff, including toastpages collection and the 32-bit dataLength overflow clamp.

  • Hans Buschmannsaw the final versionLLM analysis, 99% sureHans tested the TOAST scheduling change from #5 on Windows against a customer database and reported improved parallel dumping in #8; this patch subsequently changed only through rebasing.
Use PREPARE/EXECUTE for repetitive per-object queries in pg_dump.
6 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #140 of 1 reviewer saw the final version

The prepared-query patch in #14 matches this commit's stable patch ID and normalized diff, including PREPARE/EXECUTE conversions and connection-specific Archive.is_prepared state.

  • Andres Freundsaw an earlier versionLLM analysis, 99% sureAndres reported a function-heavy dump improving from 7.4 to 2.5 seconds and identified the static-state parallelism hazard in #9, before #13 changed the state management to a per-connection array.
Avoid per-object queries in performance-critical paths in pg_dump.
6 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1
LLM analysis, 100% surelanded as postedfrom patchset #140 of 2 reviewers saw the final version

The bulk-query patch in #14 matches this commit's stable patch ID and normalized diff, including OID-array filtering and the PostgreSQL 8.4 minimum-version checks in pg_dump.c and pg_dumpall.c.

  • Justin Pryzbysaw an earlier versionLLM analysis, 94% sureJustin questioned the OID-array string-length checks in #6, prompting explanatory comments, but the later revision #14 also changed which source-server versions are accepted.
  • Andres Freundsaw an earlier versionLLM analysis, 96% sureAndres supplied the batching idea credited in #5 and tested an alternative for pre-8.4 servers in #9, before #14 instead raised the minimum supported source version.
Postpone calls of unsafe server-side functions in pg_dump.
6 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1REL_14_STABLE14.7REL_13_STABLE13.10+2 moreAlso in: [BUG] pg_dump blocked (modified before commit)Also in: Assorted improvements in pg_dump (not this thread)
LLM analysis, 100% surelanded as postedfrom patchset #14

The unsafe-function patch in #14 has the same stable patch ID as this commit, postponing partition-expression and type-name lookups and freeing partkeydef in dumpTableSchema(). This supersedes the attribution in thread #44991 to its older October 20 submission #490568, which lacked that cleanup.

Rethink pg_dump's handling of object ACLs.
6 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1Also in: Assorted improvements in pg_dump (not this thread)
LLM analysis, 100% surelanded as postedfrom patchset #141 of 1 reviewer saw the final version

The ACL patch in #14 has the same stable patch ID as this commit, including separate pg_init_privs collection, client-side ACL comparison, and the schema-specific empty default-ACL baseline. This supersedes the attribution in thread #44991 to its older October 20 submission #490568, since the December 3 patch here already contains the committed adjustments.

  • Justin Pryzbysaw the final versionLLM analysis, 99% sureJustin reported that the ACL patch reduced single-table dump overhead from 6.9 to 1.2 seconds in #6, and the ACL patch changed only through rebasing afterward.
Refactor pg_dump's tracking of object components to be dumped.
6 Dec 2021Tom Lanemaster15beta1REL_19_STABLE15beta1REL_18_STABLE15beta1REL_17_STABLE15beta1REL_16_STABLE15beta1REL_15_STABLE15beta1Also in: Assorted improvements in pg_dump (not this thread)
LLM analysis, 100% surelanded as postedfrom patchset #14

The component-tracking patch in #14 has the same stable patch ID as this commit, including separate requested/available component masks and copied comment/security-label strings. This supersedes the attribution in thread #44991 to its older October 20 submission #490568, because the December 3 patch here matches the committed version exactly.

Credits across 7 commits

Author
Tom LaneLLM7
Committer
Tom Lane7

N LLM means N of the credits come from LLM analysis, not from commit trailers.