On 12/1/17, 11:16 AM, "Tomas Vondra" <tomas.vondra@2ndquadrant.com> wrote:
>> The behavior I would like to see is that the void ignores this table and
>> moves to another instead of being blocked.
>>
>
> I believe autovacuum should not block waiting for a heavy-weight lock on
> a table since this commit that went into 9.1:
>
> https://git.postgresql.org/gitweb/?p=postgresql.git;a=commit;h=32896c40ca766146312b28a5a0eb3f66ca0300ed
>
> So I'm wondering what problem you're running into.
It sounds like Lyes is doing a periodic database-wide VACUUM that is
getting blocked on certain relations that are already locked (perhaps
because autovacuum is already working on it). IIUC, the ask is to provide
a way to skip vacuuming relations that cannot be immediately locked.
There is already an internal flag called VACOPT_NOWAIT that gives
autovacuum this ability in certain cases (introduced in the aforementioned
commit), and I recently brought up this potential use-case as
justification for a patch [0]. I'd be happy to submit a patch for
providing VACOPT_NOWAIT to users if others think it's something we should
do.
As a side note, this seems more like a feature request than a bug report,
so I'm adding pgsql-hackers@ here, too.
Nathan
[0] https://www.postgresql.org/message-id/875815E8-7A99-4488-AA07-F254C404E2CF%40amazon.com