Skip Navigation
Lemmy Support @lemmy.ml DosDude👾 @retrolemmy.com

[SOLVED] Postgres using a lot of CPU since update

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.

20

You're viewing a single thread.

20 comments

  • Do you have any tweaks of pg settings? Eg. shm mem, shared_buffers etc. ?

    If not you migh want to: https://pgtune.leopard.in.ua/

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

      • 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.

        • 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'

          • extension "pg_stats_statements" is not available

            According to this https://stackoverflow.com/a/72966651/5881796

            The extension is not loaded: CREATE EXTENSION pg_stat_statements;

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

              # SELECT count(*) FROM pg_stat_activity;
 count
-------
    11
(1 row)

              • 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)

                • I think so. I have lemmy and everything needed running through a single docker container using Lemmy-Easy-Deploy.

                  • 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

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

                      The pgautoupgrade was added for the new version because this deployment is an all-in-one solution for running lemmy. And upgrading the databases turned out to be quite the effort until some user pointed the maintainer towards pgautoupgrade here.

                      I tried running lemmy before I found out about this, but this just makes it so much more convenient to run.

You've viewed 20 comments.