Horton

Re: Fix parallel vacuum buffer usage reporting

Anthonin Bonnefoy <anthonin.bonnefoy@datadoghq.com>

View thread
From: Anthonin Bonnefoy <anthonin.bonnefoy@datadoghq.com>
To: Alena Rybakina <lena.ribackina@yandex.ru>
Cc: Masahiko Sawada <sawada.mshk@gmail.com>, PostgreSQL Hackers <pgsql-hackers@postgresql.org>
Date: 2024-04-25T07:17:52Z
Lists: pgsql-hackers

Commits

Same data as JSON: GET /api/v1/messages/:b64id/commits the thread's linked commits as JSON, with link sources. API reference →

  1. Fix parallel vacuum buffer usage reporting.

  2. Use bump context for TID bitmaps stored by vacuum

On Wed, Apr 24, 2024 at 4:01 PM Alena Rybakina <lena.ribackina@yandex.ru>
wrote:

> I tested the main postgres branch with and without your fix using a script
> that was written by me. It consists of five scenarios and I made a
> comparison in the logs between the original version of the master branch
> and the master branch with your patch:
>
 Hi! Thanks for the tests.

I have attached a test file (vacuum_check_logs.sql)
>
Looking at the script, you won't trigger the problem. The reporting issue
will only happen if there's a parallel index vacuum and it will only happen
if there's at least 2 indexes [0]. You will need to create an additional
index.

The same script was run, but using vacuum verbose analyze, and I saw the
> difference again in the fifth step:
> with your patch: buffer usage: 32312 hits, 607 misses, 1566 dirtied
> master: buffer usage: 32346 hits, 573 misses, 1360 dirtied
>
Isn't there a chance for the checkpointer to run during this time? That
could make the conditions between the two runs slightly different and
explain the change in buffer report.

[0]
https://github.com/postgres/postgres/blob/8a1b31e6e59631807a08a4e9465134c343bbdf5e/src/backend/access/heap/vacuumlazy.c#L2826-L2831