EDIT: It seems to have been fixed thanks to @graycube@lemmy.world. Running analyze verbose; in postgres.

After updating to 0.19.5 from 0.19.3 my postgres is often using up 500%+ of CPU according to docker stats and often going to 100% CPU on most cores according to htop. Also noticed in the uptime monitor:

htop shows one of the 5 postgres processes constantly on UPDATE. I think this might be part of the problem.

I’m not comfortable with postgres and am honestly completely in the dark how or where to mitigate or even pinpoint this issue.

Any help would be appreciated.

    • DosDude👾OPA
      2·
      5 months ago

      I have. And I tried to tweak it with no avail. But it was working within acceptable levels before the update.

      • taaz@biglemmowski.winEnglish
        2·
        5 months ago

        hmm, how many connections are used SELECT count(*) FROM pg_stat_activity; ?
        I am not a master postgres admin but my intuition has been that the amount of connections is a big factor in how pg behaves with cpu and mem.

        • DosDude👾OPA
          1·
          5 months ago
          ERROR:  extension "pg_stats_statements" is not available

          Even though it’s added in the customPostgresql.conf shared_preload_libraries = 'auto_explain,pg_stat_statements'

            • DosDude👾OPA
              2·
              5 months ago

              I added pg_stat_statements, and ran it. This was the result:

              # SELECT count(*) FROM pg_stat_activity;
 count
-------
    11
(1 row)
              • taaz@biglemmowski.winEnglish
                3·
                5 months ago

                do you also have pict-rs connected to this postgres instance? that is surprisingly low number to me, I would have expected anywhere between 20-50 active connections (I use 50 for lemmy and 20 for pict-rs, configured in their respective conf files)

                  • taaz@biglemmowski.winEnglish
                    1·
                    5 months ago

                    Ah, you are using pretty different deployment then, even the used postgres image is different then the usual deployment ( pgautoupgrade/pgautoupgrade:16-alpine instead of postgres:16-alpine) this might or might not cause differences.

                    I would try increasing POSTGRES_POOL_SIZE to 10-20, but I am guessing here, the idea being that lemmy is hammering postgres through the default 5 conns which increases CPU but that is a bit of stretch