On Tue, Jun 16, 2020 at 10:01 AM Jim Hurne <jhurne@us.ibm.com> wrote:
Other than the increasing elapsed times for the autovacuum, we don't see any other indication in the logs of a problem (no error messages, etc).
We're currently using PostgreSQL version 10.10. Our service is JVM-based and we're using the PostgreSQL JDBC driver version 42.2.5.
Have we stumbled upon a potential bug here, or do we need to tweak some autovacuum settings? What should we look at next or what should we try next to further troubleshoot this?
What are your current autovacuum settings? Do you have long running transactions with any frequency? Decreasing autovacuum_vacuum_cost_delay to 1 or 2ms may be prudent (default changes from 20ms down to 2ms with PG 12). Is this a destructive queue of sorts with no rows permanently stored? If so, I would expect that a daily scheduled vacuum analyze may be the best course of action.
