Re: Workqueue performance

Jason Armstrong <ja(at)riverdrums(dot)com> writes:
> 1. NOTIFY/LISTEN was causing performance to degrade badly over time. I
> have reworked my code to poll the database instead.

FWIW, you need to ensure pg_listener gets vacuumed pretty aggressively
in order to prevent degradation in a high-traffic NOTIFY application.

PG 9.0 will have a completely rewritten LISTEN/NOTIFY implementation
that avoids use of a table and should scale a lot better, as well as not
needing vacuuming support. That doesn't help you right now, but
depending on what your development timescale is, you might want to plan
to go back to LISTEN/NOTIFY later.

regards, tom lane