On Tue, Feb 28, 2012 at 8:00 PM, Jon Nelson <jnelson+pgsql(at)jamponi(dot)net> wrote:
> I cranked the logging /all/ the way up and isolated the server.
> I suspect that your theory is correct.
Another option, depending on your SQLA version, when connections are
sent back to the pool, I seem to remember they were reset. That would
also reset the work_mem, you'd still see the same pid on PG logs, but
it's not the same session.