On 9/28/17, 12:20 AM, "Michael Paquier" <michael.paquier@gmail.com> wrote:
> On Thu, Sep 28, 2017 at 1:20 AM, Bossart, Nathan <bossartn@amazon.com> wrote:
>> On 9/26/17, 1:38 PM, "Bossart, Nathan" <bossartn@amazon.com> wrote:
>>> On 9/25/17, 12:42 AM, "Michael Paquier" <michael.paquier@gmail.com> wrote:
>>>> + if (!IsAutoVacuumWorkerProcess())
>>>> + ereport(WARNING,
>>>> + (errmsg("skipping \"%s\" --- relation no longer exists",
>>>> + relation->relname)));
>>>> I like the use of WARNING here, but we could use as well a LOG to be
>>>> consistent when a lock obtention is skipped.
>>>
>>> It looks like the LOG statement is only emitted for autovacuum, so maybe
>>> we should keep this at WARNING for consistency with the permission checks
>>> below it.
>>
>> I've left this as-is for now. I considered emitting this statement as a
>> LOG for autovacuum, but I'm not sure there is terribly much value in
>> having autovacuum explain that it is skipping a relation because it was
>> concurrently dropped. Perhaps this is something we should emit at a
>> DEBUG level. What do you think?
>
> DEBUG would be fine as well for me. Now that your patch provides a
> RangeVar consistently for all code paths, the message could show up
> unconditionally.
Alright, I've added logging for autovacuum in v23. I ended up needing to
do a little restructuring to handle the case when the relation was skipped
because the lock could not be obtained. While doing so, I became
convinced that LOG was probably the right level for autovacuum logs.
Nathan
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers