Re: Rethinking autovacuum.c memory handling

On 9/23/17, 5:27 AM, "Michael Paquier" <michael(dot)paquier(at)gmail(dot)com> wrote:
>On Sat, Sep 23, 2017 at 6:09 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>> I notice that autovacuum.c calls autovacuum_do_vac_analyze, and
>> thereby vacuum(), in TopTransactionContext. This doesn't seem
>> like a terribly great idea, because it doesn't correspond to what
>> happens during a manually-invoked vacuum.
>
> Indeed, the inconsistency is not good here.
>
>> What I think we should do instead is invoke autovacuum_do_vac_analyze
>> in the PortalContext that do_autovacuum has created, which we already
>> have a mechanism to reset once per table processed in do_autovacuum.
>>
>> The attached patch does that, and also modifies perform_work_item()
>> to use the same approach. Right now perform_work_item() has a
>> copied-and-pasted MemoryContextResetAndDeleteChildren(PortalContext)
>> call in its error recovery path, but that seems a bit out of place
>> given that perform_work_item() isn't using PortalContext otherwise.
>
> I have spent some time looking at your patch and testing it. This
> looks sane. A small comment that I have would be to add an assertion
> at the top of perform_work_item to be sure that it is called in the
> memory context of AutovacMemCxt.

This looks reasonable to me as well. I haven't noticed any issues after
a couple hours of pgbench with aggressive autovacuum settings, either.

Nathan