Re: How does max_parallel_workers_per_gather change load averages?

From: David Rowley <david(dot)rowley(at)2ndquadrant(dot)com>
To: Ben Nachtrieb <bnachtrieb(at)crescat(dot)net>
Cc: "pgsql-performance(at)postgresql(dot)org" <pgsql-performance(at)postgresql(dot)org>
Subject: Re: How does max_parallel_workers_per_gather change load averages?
Date: 2017-10-03 21:44:16
Message-ID: CAKJS1f8jDOsdvSNZzZn4ocXReHRC+U8D3JaHU7h0x3RhpEdq3Q@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-performance

On 4 October 2017 at 08:48, Ben Nachtrieb <bnachtrieb(at)crescat(dot)net> wrote:
> I have 2 cores and my max_parallel_workers_per_gather = 2 and
> max_worker_processes = 8, but my load averages are between 8 and 5 with
> scheduled at 1/189 to 5/195. Are these so high because I increased
> max_parallel_workers_per_gather? My understanding is that if my load
> averages are greater than my number of cores the system is overloaded.
> Should I think about it differently once I increase
> max_parallel_workers_per_gather? How should I think about it?

Parallel query is not 100% efficient. For example, adding twice the
CPU, in theory, will never double the performance, there's always some
overhead to this. It's really only useful to do on systems with spare
CPU cycles to perform this extra work. You don't seem to have much to
spare, so you may get along better if you disable parallel query.

In response to

Browse pgsql-performance by date

  From Date Subject
Next Message Jim Nasby 2017-10-08 19:25:54 Regression from 9.4-9.6
Previous Message Mark Kirkwood 2017-10-03 21:16:37 Re: select with max functions