Hi,
On 2020-09-08 17:36:00 -0300, Alvaro Herrera wrote:
> On 2020-Sep-08, Andres Freund wrote:
>
> > That made me wonder if it'd be worthwhile to add an option that'd make
> > user invoked VACUUM be interruptible by conflicting lock requests, just
> > like autovacuum is.
>
> Yeah, I recall a request for this in the past, too.
>
> > So how about adding an INTERRUPTIBLE option to VACUUM and ANALYZE?
>
> +1 on adding it to VACUUM. I'm not sure about ANALYZE ... most of the
> time it is not as bad as VACUUM in terms of blocking other things, and
> things get ugly if that ANALYZE is part of a transaction. I think I'd
> leave that out, since we don't have to cover all DDL that could take
> ShareUpdateExclusive lock (CIC etc). VACUUM is especially problematic,
> ISTM, which is we would do it for that.
ANALYZE sometimes is worse than VACUUM in some ways, it'll often take
longer on large tables (more random IO, no freeze map equivalent). And
there's still TRUNCATE / DROP TABLE etc that you want to be able to
interrupt (auto-)analyze.
I'm not sure it's really necessary to distinguish transaction /
non-transactional for ANALYZE. We continue to wait for the lock, right?
So signalling the transaction is fine, it'll error out. Even in the
case of ANALYZE in a subtransaction, we'll not do something incorrect,
we might just wait, if the top level transaction also has a lock, and
doesn't abort immediately.
So it doesn't seem that useful to not make manual analyze interruptible?
Greetings,
Andres Freund