Mar 2025

3 commits
Fix test name and username used in failed connection attempts
9 Mar 2025Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #38

Both changed lines in src/test/postmaster/t/002_connection_limits.pl match #38. Only surrounding context changes because connect_fails_wait() landed first.

tests: Fix race condition in postmaster/002_connection_limits
7 Mar 2025Andres Freundmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #391 of 1 reviewer saw the final version

The connect_fails_wait() helper and its three call-site conversions in src/test/postmaster/t/002_connection_limits.pl exactly match the corresponding patch in #39 by stable patch ID.

  • Heikki Linnakangassaw the final versiontrailerHeikki endorsed waiting for the backend-exit log message in #38, and the implementation remained unchanged in #39.
tests: Don't fail due to high default timeout in postmaster/003_start_stop
7 Mar 2025Andres Freundmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #391 of 1 reviewer saw the final version

The four-line authentication_timeout cap in src/test/postmaster/t/003_start_stop.pl has the same stable patch ID as the corresponding patch in #39.

  • Heikki Linnakangassaw the final versiontrailerHeikki explicitly approved all four patches in #38, and #39 changed their commit descriptions without changing the code.

Nov 2024

4 commits
Pass MyPMChildSlot as an explicit argument to child process
14 Nov 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #241 of 1 reviewer saw the final version

Matches the explicit child_slot argument patch in #24, with an additional explanatory comment above postmaster_child_launch() in src/backend/postmaster/launch_backend.c.

  • Andres Freundsaw the final versiontrailerAndres reviewed the remaining child-process series in #9; subsequent changes to explicit slot passing were adaptation to preceding patches and documentation.
Assign a child slot to every postmaster child process
14 Nov 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #240 of 1 reviewer saw the final version

Retains the PMChildPool implementation from #24, but adds bootstrap and single-user initialization, inlines assign_backendlist_entry(), simplifies canAcceptConnections(), and renames slot-management functions. These final edits are described in #25.

  • Andres Freundsaw an earlier versiontrailerAndres challenged pool allocation, duplicated sizing and shutdown handling in #9, #13 and #15, before the pool rewrite and final initialization fixes.
Kill dead-end children when there's nothing else left
14 Nov 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #240 of 1 reviewer saw the final version

The shutdown signaling and Cluster.pm timeout support match #24; differences are wording changes in the new start/stop test's explanatory comments.

  • Andres Freundsaw an earlier versiontrailerAndres reviewed dead-end shutdown and objected to hardcoded test timeouts in #9, before signaling changes and the expanded shutdown test in #24.
Replace postmaster.c's own backend type codes with BackendType
14 Nov 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #240 of 1 reviewer saw the final version

Matches the BackendType and BackendTypeMask conversion in #24, apart from clarifying that CountChildren() excludes special children.

  • Andres Freundsaw an earlier versiontrailerAndres identified stale type references and requested safer bitmask conversion in #13, before BackendTypeMask and its helper functions were introduced.

Oct 2024

3 commits
Use an shmem_exit callback to remove backend from PMChildFlags on exit
8 Oct 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #61 of 1 reviewer saw the final version

Implements the shmem-exit callback patch from #6, renaming MarkPostmasterChildActive() to RegisterPostmasterChildActive() and documenting hook registration in src/backend/storage/ipc/pmsignal.c.

  • Andres Freundsaw the final versiontrailerAndres requested a registration-oriented function name because the operation installs an exit hook in #9, and that naming change was adopted.
Add test for dead-end backends
8 Oct 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #220 of 1 reviewer saw the final version

The raw_connect()/raw_connect_works() additions and dead-end connection tests have the same stable patch ID as the second patch in #22.

  • Andres Freundsaw an earlier versiontrailerAndres questioned the need for raw sockets in #9, before SSLRequest synchronization and the raw_connect_works() capability probe were added.
Add test for connection limits
8 Oct 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #221 of 1 reviewer saw the final version

The new src/test/postmaster suite and connection-limit test exactly match the first patch in #22; its stable patch ID is unchanged across the earlier test submissions.

  • Andres Freundsaw the final versiontrailerAndres reviewed the test additions and their README instructions in #9, and this connection-limit patch remained unchanged through #22.

Aug 2024

7 commits
Fix garbled process name on backend crash
19 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surewritten from the discussion, no matching patch1 of 1 reviewer saw the final version

Changes HandleChildCrash()'s argument from namebuf to procname in CleanupBackend(), directly fixing Alexander Lakhin's uninitialized-name diagnosis in #7. No attachment contains this fix; Heikki confirms it in #8.

  • Alexander Lakhinsaw the final versionLLM analysis, 98% sureAlexander reported garbled crash logs and identified the potentially uninitialized namebuf in #7, directly motivating the one-line correction.
Consolidate postmaster code to launch background processes
12 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitlight changesfrom patchset #31 of 1 reviewer saw the final version

Preserves the LaunchMissingBackgroundProcesses() extraction from #3, with rewritten comments explaining process lifetimes and startup conditions. Heikki describes the comment cleanup in #5 and confirms landing in #6.

  • Thomas Munrosaw the final versiontrailerThomas explicitly endorsed removing redundant launches and extracting the ServerLoop helper in #4, after which only comments changed.
Fix comment on processes being kept over a restart
9 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #31 of 1 reviewer saw the final version

The postmaster restart-comment correction has the same stable patch ID as the corresponding patch in #3.

  • Thomas Munrosaw the final versiontrailerThomas approved the initial refactoring and restart-comment patches in #4, and this patch landed unchanged.
Refactor code to handle death of a backend or bgworker in postmaster
9 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% suremodified before commitmoderate changesfrom patchset #30 of 1 reviewer saw the final version

Uses the unified CleanupBackend() implementation from #3, adding a logged flag and conditional logging to avoid duplicate ERROR_WAIT_NO_CHILDREN messages. The local logic adjustment and comment edits are acknowledged in #5.

  • Thomas Munrosaw an earlier versiontrailerThomas examined unified cleanup and Windows exit-status handling in #4, before the duplicate-log suppression logic described in #5 was added.
Make BackgroundWorkerList doubly-linked
9 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #31 of 1 reviewer saw the final version

The BackgroundWorkerList conversion and worker-pointer interfaces exactly match the first patch in #3 by stable patch ID.

  • Thomas Munrosaw the final versiontrailerThomas approved the doubly linked list prerequisite in #4, and it landed unchanged.
Minor refactoring of assign_backendlist_entry()
1 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2

The assign_backendlist_entry() allocation/interface refactor exactly matches the second patch in #2 by stable patch ID; #3 announces its landing.

Fix outdated comment; all running bgworkers are in BackendList
1 Aug 2024Heikki Linnakangasmaster18beta1REL_19_STABLE18beta1REL_18_STABLE18beta1
LLM analysis, 100% surelanded as postedfrom patchset #2

The comment correction in src/include/postmaster/bgworker_internals.h exactly matches the first patch in #2 by stable patch ID; #3 announces its landing.

Credits across 17 commits

Author
Heikki LinnakangasLLM14
Andres FreundLLM2
Reviewer
Andres Freund7
Thomas Munro4
Heikki Linnakangas2
Tested by
Tomas Vondra1
Reported by
Alexander Lakhin1
Tomas Vondra1
Committer
Heikki Linnakangas15
Andres Freund2

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