Re: Logical decoding CPU-bound w/ large number of tables

Mathieu Fenniak <mathieu(dot)fenniak(at)replicon(dot)com> writes:
> I've taken your patches, Tom, and applied them to a test deployment of my
> actual application.

> The most accessible way I have to reproduce this issue is to run a
> maintenance task that we typically run during a software deployment, which
> will remove some tenant schemas from our database system. I ran that task
> with and without the patches, and monitored how long logical replication
> stalled. This test was on a database with 22 schemas containing 400 tables
> each (~8800 tables).

> As a baseline, PostgreSQL 9.5.6 stalled completely for 21m 13s

> PostgreSQL 9.5.6 + all three patches stalled completely for 4m 11s.

> Definitely a fantastic improvement. :-)

Thanks for confirming that these patches help in a real-world scenario.
Obviously there's still room for improvement, but I think it's worth
pushing what I've done, so I'll go do that. I hope Andres will follow
up with trying to improve matters from the other angle.

regards, tom lane