Hi Vlad,
Concurrency problems are the most difficult to discover and
reproduce. Thanks for that work!
Since every patch is subject to testing, it is recommended to create a
CommitFest card for it. Once in the CF-card, the patch set is picked up
by CFBot and is thoroughly tested. You also will see whether the patch
successfully applies to the target branch or not.
You also might want to stick to a single PostgreSQL version in a thread,
as all attached files are applied to the upstream during a CFBot job,
which is impossible when a mail message includes patches for several
PG versions simultaneously. Adhering to a single version per a thread
and using a CommitFest card also make reviewers' task easier, for the
reason that they can see the version and the base commit to which the
patch should be applied. Patches for other PG's versions can probably
be moved to separated threads.
It is all by now. I am still looking into the code of the patch. I
hope, later I can give some useful comments in regard to the solution
itself.
P.s. I tried to apply the patch to 3b28dad70e2. After
0002-ProcKill-PG18-core-lockgroup-freelist-race had been applied
successfully, subsequent applying of
0003-PG18-unfixed-repro-tap-injection-harness failed.
Best regards,
Evgeny Voropaev,
Tantor Labs, LLC.
