On Wed, 5 May 2004, Chris Gamache wrote:
> When I run an query (that I have identified as expensive, and non-critical), I
> would like to be able to take its priority down a few notches to allow the
> standard every-day tasks of the DB to run unhindered. As it stands, postmasters
> elbow each other out for processor time, and it seems like one expensive query
> can bring other DB functions to a crawl.
>
> Hardware : Dual P4 Xeon 2.8GHz; 1 GB RAM; ULTRA320 RAID 10 with an ext3
> filesystem
> PostgreSQL 7.4.2
>
> What can be done to allow for smarter preemption? Could I do anything at the OS
> level to throttle that particular postmaster's process? I'm running (IMO) a
> balanced config, but there's always room for improvement. Its that oddball
> query that comes around once every so often that causes the problem.
It is inadvisable to change priority of backends as that could lead to
deadlocks in certain situations I believe.
Have you read:
http://www.varlena.com/varlena/GeneralBits/Tidbits/perf.html
yet? It may just be that you don't have enough shared_buffers or whatnot.