So I did an upgrade today and the postgresql version seems to have been updated in the process. Since doing this, this issue (or similar) has returned (old topic is closed now): [image] Very slow Sidekiq issue with large queue due to massive numbers of unread user notificatio…

Ok so to follow up, everything still looks good, postmaster has calmed down drastically and sidekiq is moving really fast again. To summarise on what worked in my scenario (with default db name of ‘discourse’): cd /var/discourse/ ./launcher enter app sudo -u postgres psql \c discourse Then from t…

[image] markersocial: VACUUM FULL VERBOSE; Can you try a vacuum verbose analyze? Also applies to you @eboehnisch . This following very simple query averaging 2s is really weird: [image] markersocial: SELECT "posts"."id" FROM "posts" WHERE ("posts"."deleted_at" IS NULL) AND "posts"."topic_…

Tried VACUUM VERBOSE ANALYZE;. What shall I be looking for in the results?

Great, thanks! So I think this has solved the issue with postmasters going wild using all the CPU. So this is a big improvement. Sidekiq is still slow unfortunately, but better than before running VACUUM VERBOSE ANALYZE;

[image] eboehnisch: What shall I be looking for in the results? Nothing special, unless it is a big error message. Performance should recover now. [image] markersocial: Great, thanks! So I think this has solved the issue with postmasters going wild using all the CPU. So this is a big im…

This improved the situation definitely but after a while again three postmaster processes use 100% CPU for over a minute each.

I’d say wait a bit, because you probably have a queue to chew from the slow period. Check /sidekiq for stats.

:smiley: Speed has really picked up now, looks like it’s fixed. Thanks a lot for the help @Falco ! I’ll leave it running for a while and post an update here later~

Maybe try running: VACUUM FULL VERBOSE; and (assuming your db name is ‘discourse’ which is default): REINDEX DATABASE discourse; This might be better than normal reindex, but I haven’t tested it: PostgreSQL 12 update

Both operations listed above generates locks that can be really hard in a live site. I’d recommend the ones listed at post-update-optional-tasks because other than using lots of CPU, they don’t block the normal database operations.

Run both. The latter reports an error, however: discourse=# REINDEX SCHEMA CONCURRENTLY public; ERROR: could not create unique index "index_users_on_username_lower_ccnew" DETAIL: Key (username_lower)=(marks) is duplicated. I guess it’s time to get into PostgreSQL and to find and delete the recor…

Slow Sidekiq + Postmaster using 95%+ CPU (32 cores) after Postgresql Version Upgrade

Support Self-hosting

sam (Sam Saffron) May 28, 2020, 12:58am 18

see:

Topic		Replies	Views
Very slow Sidekiq issue with large queue due to massive numbers of unread user notifications Self-hosting	68	7698	April 11, 2020
Long-Running Sidekiq Jobs Feature	21	1777	December 24, 2020
PostgreSQL runaway IO Self-hosting	35	3176	March 3, 2021
PostgreSQL 13 update Announcements	2	11174	September 13, 2023
Primary Postgres database process (postmaster) eating all CPU Self-hosting server-resources	13	4194	May 28, 2020

Slow Sidekiq + Postmaster using 95%+ CPU (32 cores) after Postgresql Version Upgrade

Related topics