Hi,
we are running Docker based Discourse 2.7.7 with a very large user base. After upgrading from 2.4.2 to 2.7.7, we are experiencing that some Postgres process are stuck at 100% CPU. Further investigation results that such queries seem to cause the problem:
discourse_prod=# select pid, datid, query from pg_stat_activity WHERE pid = '244906';
pid | datid | query
--------+---------+-------------------------------------------------------
244906 | 2068583 | DELETE FROM user_badges +
| | WHERE id IN ( +
| | SELECT ub.id +
| | FROM user_badges ub +
| | LEFT JOIN ( +
| | SELECT id user_id, current_timestamp granted_at+
| | FROM users +
| | WHERE id IN ( +
| | SELECT p1.user_id +
| | FROM post_custom_fields pc +
| | JOIN badge_posts p1 ON p1.id = pc.post_id +
| | JOIN topics t1 ON p1.topic_id = t1.id +
| | WHERE p1.user_id <> t1.user_id AND +
| | name = 'is_accepted_answer' AND +
| | p1.user_id IN ( +
| | SELECT user_id +
| | FROM posts +
| | WHERE TRUE OR p1.id IN (-1) +
| | ) +
| | GROUP BY p1.user_id +
| | HAVING COUNT(*) > 9 +
| | ) +
| | ) q ON q.user_id = ub.user_id +
| | +
| | WHERE ub.badge_id = 103 AND q.user_id IS NULL +
| | ) +
| |
(1 row)
We found this query which, I assume, will cause a huge overhead, doesn’t it?
| | SELECT user_id +
| | FROM posts +
| | WHERE TRUE OR p1.id IN (-1) +
Any suggestions on this?
Let me know if you need further info.
Thanks,
Daniel.