RE: [Patch] Optimize dropping of relation buffers using dlist

On Wednesday, September 2, 2020 5:49 PM, Amit Kapila wrote:
> On Wed, Sep 2, 2020 at 9:17 AM Tom Lane  wrote:
> >
> > Amit Kapila  writes:
> > > Even if the relation is locked, background processes like
> > > checkpointer can still touch the relation which might cause
> > > problems. Consider a case where we extend the relation but didn't
> > > flush the newly added pages. Now during truncate operation,
> > > checkpointer can still flush those pages which can cause trouble for
> > > truncate. But, I think in the recovery path such cases won't cause a
> problem.
> >
> > I wouldn't count on that staying true ...
> >
> >
> https://www.postgresql.org/message-id/CA+hUKGJ8NRsqgkZEnsnRc2MFR
> OBV-jC
> > nacbYvtpptK2A9YYp9Q@mail.gmail.com
> >
> 
> I don't think that proposal will matter after commit c5315f4f44 because we are
> caching the size/blocks for recovery while doing extend (smgrextend). In the
> above scenario, we would have cached the blocks which will be used at later
> point of time.
> 

Hi,

I'm guessing we can still pursue this idea of improving the recovery path first.

I'm working on an updated patch version, because the CFBot's telling
that postgres fails to build (one of the recovery TAP tests fails).
I'm still working on refactoring my patch, but have yet to find a proper solution at the moment.
So I'm going to continue my investigation.

Attached is an updated WIP patch.
I'd appreciate if you could take a look at the patch as well.

In addition, attached also are the regression logs for the failure and other logs
Accompanying it: wal_optimize_node_minimal and wal_optimize_node_replica.

The failures stated in my session was:
t/018_wal_optimize.pl ................ 18/34 Bailout called.
Further testing stopped:  pg_ctl start failed
FAILED--Further testing stopped: pg_ctl start failed

Best regards,
Kirk Jamison