Re: BUG #17724: All autovacuum workers operate on 1 db at a time

On Fri, Dec 16, 2022 at 2:13 PM PG Bug reporting form <
noreply(at)postgresql(dot)org> wrote:

> The following bug has been logged on the website:
>
> Bug reference: 17724
> Logged by: Peter Plachta
> Email address: pplachta(at)gmail(dot)com
> PostgreSQL version: Unsupported/Unknown
> Operating system: Windows (Azure)
> Description:
>
> Hi all
>
> I have observed this on PG 9.6 and 10 and could not find a bug indicating
> it
> was fixed.
>
> Issue is that autovacuum workers appear bound to a certain DB, even if
> there
> is more urgent (wraparound) work to do elsewhere.
>
> In our case specifically we have multiple DBs, one has a large table that
> for various reasons takes a long time (many hours) to vacuum. During this
> time other autovacuum workers are idle while our other DBs which do
> frequent
> insert/delete accumulate significant bloat which ends up affecting query
> time.
>
> Ideally the idle autovacuum workers would just proceed to the next DB.
>
>
This doesn't qualify as a bug report. If you need assistance diagnosing
issues on your system you should use the -general mailing list, or maybe
visit a forum like Slack or Discord where once you find someone willing to
help, communications tend to be much more interactive/timely in nature.

I would suggest you upgrade to a supported release in any case.

As for diagnostics, the system is capable of producing logs, and also has
views, regarding the autovacuum daemon. You should share this raw data
instead of simply giving your interpretation of what you saw.

Sharing the relevant configuration setting values for the autovacuum daemon
is also recommended.

David J.