Re: how to investigate GIN fast updates and cleanup cycles?

Steve Kehlet <steve(dot)kehlet(at)gmail(dot)com> writes:
> On Fri, Aug 28, 2015 at 10:42 AM Jeff Janes <jeff(dot)janes(at)gmail(dot)com> wrote:
>> Note that a manual ANALYZE will *not* clear the pending list, it has to be
>> an autoanalyze.

> This is a brain bender, I didn't know there were differences, and this eats
> away a little bit at my confidence in understand things, but I'll just
> accept it for now.

A look at the source code says Jeff's right: either manual or auto VACUUM
will clear the list, and so will auto ANALYZE, but not manual ANALYZE.
Not sure why we did that, but it's clearly intentional.

>> (Manual VACUUM will clear the pending list, but you might have trouble
>> getting manual VACUUM to complete fast enough)

> You are exactly right the manual VACUUM is taking forever.

It looks like a VACUUM will do the cleanup during the first ginbulkdelete
call, so you could probably handle this by running a manual "VACUUM
VERBOSE" with the smallest possible maintenance_work_mem, and canceling it
as soon as you see something reported about the GIN index.

regards, tom lane