On Oct 3, 2012, at 11:50 AM, Tom Lane wrote:
> Ben Chobot <bench@silentmedia.com> writes:
>> 4. What might cause autovacuum analyze to make an index perform worse immediately, when a manual vacuum analyze does
nothave the same affect? And I'm not talking about changing things so the planner doesn't use the index, but rather,
havingthe index actually take longer.
>
> Dunno about the replication angle, but would this have been a GIN index?
> I'm wondering about possible interference with flushing of its
> pending-insert queue (the FASTUPDATE stuff).
Nope, btree:
create index get_delayed_jobs_index on delayed_jobs (priority, run_at) tablespace data1 where locked_at is null and
queue='queue'and next_in_strand=true;
There are half a dozen other indices on this table too (that weren't applicable to the long query) but they're all
btrees.