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

On Thursday, December 10, 2020 12:27 PM, Amit Kapila wrote:
> On Thu, Dec 10, 2020 at 7:11 AM Kyotaro Horiguchi
> <horikyota(dot)ntt(at)gmail(dot)com> wrote:
> >
> > At Wed, 9 Dec 2020 16:27:30 +0530, Amit Kapila
> > <amit(dot)kapila16(at)gmail(dot)com> wrote in
> > > On Wed, Dec 9, 2020 at 6:32 AM Kyotaro Horiguchi
> > > > Mmm. At least btree doesn't need to call smgrnblocks except at
> > > > expansion, so we cannot get to the optimized path in major cases
> > > > of truncation involving btree (and/or maybe other indexes).
> > > >
> > >
> > > AFAICS, btree insert should call smgrnblocks via
> > >
> btree_xlog_insert->XLogReadBufferForRedo->XLogReadBufferForRedoExte
> nded->XLogReadBufferExtended->smgrnblocks.
> > > Similarly delete should also call smgrnblocks. Can you be bit more
> > > specific related to the btree case you have in mind?
> >
> > Oh, sorry. I wrongly looked to non-recovery path. smgrnblocks is
> > called during buffer loading while recovery. So, smgrnblock is called
> > for indexes if any update happens on the heap relation.
> >
>
> Okay, so this means that we can get the benefit of optimization in many cases
> in the Truncate code path as well even if we use 'cached'
> flag? If so, then I would prefer to keep the code consistent for both vacuum
> and truncate recovery code path.

Yes, I have tested that optimization works for index relations.

I have attached the V34, following the conditions that we use "cached" flag
for both DropRelFileNodesBuffers() and DropRelFileNodesBuffers() for
consistency.
I added comment in 0004 the limitation of optimization when there are TOAST
relations that use NON-PLAIN strategy. i.e. The optimization works if the data
types used are integers, OID, bytea, etc. But for TOAST-able data types like text,
the optimization will be skipped and force a full scan during recovery.

Regards,
Kirk Jamison