| From: | Peter Geoghegan <pg(at)bowt(dot)ie> |
|---|---|
| To: | PostgreSQL Hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org> |
| Subject: | Buffer pin leak in hash index bucket splits |
| Date: | 2026-10-11 04:41:36 |
| Message-ID: | CAH2-Wz=b9sSBv15HsgZAuazjJu8-YRGHOpLjvp+dJgEDaBx0Ng@mail.gmail.com |
| Views: | Whole Thread | Raw Message | Download mbox | Resend email |
| Thread: | |
| Lists: | pgsql-hackers |
During review of what became bugfix commit 173cf85a, Claude Code found
an unrelated bug in hash index bucket splits.
Attached bug fix patch adds an isolation test demonstrating the
problem; with the fix itself reverted, running the Claude-authored
test shows a buffer pin leak. The leak is caused by _hash_finish_split
neglecting to release a pin on the primary bucket buffer in its
failure paths, used when a cleanup lock cannot be acquired. The fix
just has _hash_finish_split release the pin in those paths.
I doubt that it's worth committing the test itself. It'll add test
cycles for a trivial issue, that's unlikely to ever regress in the
future.
--
Peter Geoghegan
| Attachment | Content-Type | Size |
|---|---|---|
| v1-0001-Fix-hash-bucket-split-buffer-pin-leak.patch | application/octet-stream | 6.9 KB |
| From | Date | Subject | |
|---|---|---|---|
| Next Message | Amit Kapila | 2026-10-11 05:14:36 | Re: Fix WITHOUT OVERLAPS multirange with location replication |
| Previous Message | kedar anavardekar | 2026-10-11 04:31:06 | Re: Fix WITHOUT OVERLAPS multirange with location replication |