Igor Polishchuk <ipolishchuk@hi5.com> writes:
> ... Eventually, the database just hangs and the host
> becomes unresponsive for about 15 minutes till the sessions die out.
Well, that's a problem all right, but I'm entirely unconvinced by your
hypothesis about what's causing it. Have you tried watching "vmstat"
(particularly the swap rate) while things go downhill? Do any signs
of distress start to appear in the postmaster log or other system logs?
Do other services on the box start to fail, and if so what are they
complaining about?
> Isn't it true, that work memory once allocated for a session does not get
> deallocated till the sessions is closed?
No.
regards, tom lane