Apr 2019

2 commits
Avoid "could not reattach" by providing space for concurrent allocation.
9 Apr 2019Noah Mischmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+6 more
LLM analysis, 100% surelanded as postedfrom patchset #291 of 1 reviewer saw the final version

Applying the substantive hunks from #29 to the commit's parent produces identical committed contents in win32_shmem.c, postmaster.c and pg_shmem.h; only postmaster.c line offsets changed. The attachment's signal.c and open.c sections contain no hunks.

  • Tom Lanesaw the final versionLLM analysis, 100% sureTom identified socket-lifetime problems in the retry prototype at #23, challenged the final patch's allocator assumption at #30, and is explicitly credited as reviewer in 617dc6d299.
Assert that pgwin32_signal_initialize() has been called early enough.
4 Apr 2019Noah Mischmaster12beta1REL_19_STABLE12beta1REL_18_STABLE12beta1REL_17_STABLE12beta1REL_16_STABLE12beta1REL_15_STABLE12beta1REL_14_STABLE12beta1REL_13_STABLE12beta1+1 more
LLM analysis, 100% surewritten from the discussion, no matching patch

#29 explains that pg_usleep() did nothing before signal initialization and explicitly promises assertions. This commit adds those assertions in src/backend/port/win32/signal.c and src/port/open.c; the attachment contains only empty diff headers for those files.

Apr - May 2018

9 commits
Remove investigative code for can't-reattach-to-shared-memory errors.
1 May 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 100% surewritten from the discussion, no matching patch

This restores src/backend/port/win32_shmem.c to its pre-investigation contents, implementing Tom's announced removal of unsuccessful diagnostic experiments in #21. It predates both attached patchsets.

Does it help to wait before reattaching?
1 May 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 100% surewritten from the discussion, no matching patch

#15 announces a one-second sleep experiment after the preliminary map/unmap attempt failed. The commit removes that attempt and adds pg_usleep(1000000L) before VirtualFree in PGSharedMemoryReAttach().

Map and unmap the shared memory block before risking VirtualFree.
30 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 100% surewritten from the discussion, no matching patch

The preliminary MapViewOfFileEx(NULL)/UnmapViewOfFile sequence in win32_shmem.c implements Tom's proposal in #10 to trigger infrastructure allocations before releasing the reserved address range. No matching attachment was posted.

Further effort at preventing memory map dump from affecting the results.
30 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 100% surewritten from the discussion, no matching patch

#14 proposes capturing memory maps into previously allocated storage to avoid logging-induced allocations. The commit changes dumpmem() to append into a preallocated StringInfo and defers logging until mapping fails.

Remove Windows module-list-dumping code.
30 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 100% surewritten from the discussion, no matching patch

#10 reports that module enumeration greatly increases failures and should be removed. The commit deletes dumpdlls(), removes PSAPI build dependencies and retains memory-map probing through VirtualQuery.

Dump full memory maps around failing Windows reattach code.
30 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 99% surewritten from the discussion, no matching patch

This replaces narrow probes with full memory/module dumps and PSAPI build support, drawing on Noah's earlier diagnostic implementation linked in #4. Tom confirms integrating that code and describes its effects in #10; neither attached patchset contains it.

Get still more info about Windows can't-reattach-to-shared-memory errors.
30 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 99% surewritten from the discussion, no matching patch

This extends the investigation introduced in #8 by probing immediately after VirtualFree and capturing post-failure state before logging. Its commit message ties these unattached win32_shmem.c changes to this thread's suspected allocation race.

Get more info about Windows can't-reattach-to-shared-memory errors.
29 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 99% surewritten from the discussion, no matching patch

This adds a pre-VirtualFree VirtualQuery probe and corresponding failure diagnostics in PGSharedMemoryReAttach(), extending the investigation described in #8. It predates both attached patchsets and its Discussion trailer identifies this thread.

Try to get some info about Windows can't-reattach-to-shared-memory errors.
28 Apr 2018Tom Lanemaster11beta1REL_19_STABLE11beta1REL_18_STABLE11beta1REL_17_STABLE11beta1REL_16_STABLE11beta1REL_15_STABLE11beta1REL_14_STABLE11beta1REL_13_STABLE11beta1+2 more
LLM analysis, 100% surewritten from the discussion, no matching patch

#8 announces this diagnostic push to investigate allocation-size rounding. The commit adds mapped-size logging in PGSharedMemoryCreate() and VirtualQuery diagnostics after failed reattachment, without an attached patch.

Credits across 11 commits

Author
Noah MischLLM1
Committer
Tom Lane9
Noah Misch2

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