Re: subscription worker signalling wal writer too much

From: Jeff Janes <jeff(dot)janes(at)gmail(dot)com>
To: pgsql-hackers <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: subscription worker signalling wal writer too much
Date: 2017-06-14 22:08:49
Message-ID: CAMkU=1yz3RYqPta7kzv0dyPQNaGoutkYKmTWBLFKpsPc1DO9nQ@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox
Thread:
Lists: pgsql-hackers

On Wed, Jun 14, 2017 at 11:55 AM, Jeff Janes <jeff(dot)janes(at)gmail(dot)com> wrote:

> If I publish a pgbench workload and subscribe to it, the subscription
> worker is signalling the wal writer thousands of times a second, once for
> every async commit. This has a noticeable performance cost.
>

I've used a local variable to avoid waking up the wal writer more than once
for the same page boundary. This reduces the number of wake-ups by about
7/8.

I'm testing it by doing 1e6 transactions over 8 clients while replication
is in effect, then waiting for the logical replica to catch up. This cycle
takes 183.1 seconds in HEAD, and 162.4 seconds with the attached patch.
N=14, p-value for difference of the means 6e-17.

If I suppress all wake-ups just to see what would happen, it further
reduces the runtime to 153.7.

Cheers,

Jeff

Attachment Content-Type Size
wake_wal_writer_less_aggressively.patch application/octet-stream 877 bytes

In response to

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Andres Freund 2017-06-14 22:20:19 Re: subscription worker signalling wal writer too much
Previous Message Tom Lane 2017-06-14 22:01:58 Re: intermittent failures in Cygwin from select_parallel tests