If the problem occurs gradually (like leaking 20 connections per hour during ETL), you can check pg_stat_activity every hour or so to try to observe the problem before all the connection slots are used up, to collect diagnostic information. Alternately, leave a connection opened to the DB and wait until all connection slots *are* used up, and then check pg_stat_activity. That will take longer, and you'll have more information to weed through. What messages are in the server's log ? v11.2 is years old and hundreds of bugfixes behind. Since you ran into this problem anyway, why not run 11.16, which was released today ? How did you install postgres 11 ? From source or from packages ? Which packages ? The semi-official PGDG RPM packages are available here: https://yum.postgresql.org/ -- Justin
