Date: Thu, 19 Mar 2020 15:44:49 -0700
From: Andres Freund <andres@anarazel.de>
To: pgsql-hackers@postgresql.org
Subject: Why does [auto-]vacuum delay not report a wait event?
Message-ID: <20200319224449.pbjrivflbf4x76tj@alap3.anarazel.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Precedence: bulk

Hi,

I was looking at [1], wanting to suggest a query to monitor what
autovacuum is mostly waiting on. Partially to figure out whether it's
mostly autovacuum cost limiting.

But uh, unfortunately the vacuum delay code just sleeps without setting
a wait event:

void
vacuum_delay_point(void)
{
...
	/* Nap if appropriate */
	if (msec > 0)
	{
		if (msec > VacuumCostDelay * 4)
			msec = VacuumCostDelay * 4;

		pg_usleep((long) (msec * 1000));


Seems like it should instead use a new wait event in the PG_WAIT_TIMEOUT
class?

Given how frequently we run into trouble with [auto]vacuum throttling
being a problem, and there not being any way to monitor that currently,
that seems like it'd be a significant improvement, given the effort?


It'd probably also be helpful to report the total time [auto]vacuum
spent being delayed for vacuum verbose/autovacuum logging, but imo
that'd be a parallel feature to a wait event, not a replacement.


Greetings,

Andres Freund

[1] https://postgr.es/m/CAE39h22zPLrkH17GrkDgAYL3kbjvySYD1io%2BrtnAUFnaJJVS4g%40mail.gmail.com