Re: Improve eviction algorithm in ReorderBuffer

On Mon, Apr 1, 2024 at 11:26 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
>
> On Fri, Mar 29, 2024 at 7:37 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
> >
> > On Fri, Mar 29, 2024 at 12:13 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> > >
> > > On Fri, Mar 29, 2024 at 2:09 PM vignesh C <vignesh21@gmail.com> wrote:
> > > >
> > > > On Tue, 26 Mar 2024 at 10:05, Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> > > > >
> > > > > On Thu, Mar 14, 2024 at 12:02 PM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> > > > > >
> > > > > >
> > > > > > I've attached new version patches.
> > > > >
> > > > > Since the previous patch conflicts with the current HEAD, I've
> > > > > attached the rebased patches.
> > > >
> > > > Thanks for the updated patch.
> > > > One comment:
> > > > I felt we can mention the improvement where we update memory
> > > > accounting info at transaction level instead of per change level which
> > > > is done in ReorderBufferCleanupTXN, ReorderBufferTruncateTXN, and
> > > > ReorderBufferSerializeTXN also in the commit message:
> > >
> > > Agreed.
> > >
> > > I think the patch is in good shape. I'll push the patch with the
> > > suggestion next week, barring any objections.
> > >
> >
> > Few minor comments:
> > 1.
> > @@ -3636,6 +3801,8 @@ ReorderBufferCheckMemoryLimit(ReorderBuffer *rb)
> >   Assert(txn->nentries_mem == 0);
> >   }
> >
> > + ReorderBufferMaybeResetMaxHeap(rb);
> > +
> >
> > Can we write a comment about why this reset is required here?
> > Otherwise, the reason is not apparent.
>
> Yes, added.
>
> >
> > 2.
> > Although using max-heap to select the largest
> > + * transaction is effective when there are many transactions being decoded,
> > + * there is generally no need to use it as long as all transactions being
> > + * decoded are top-level transactions. Therefore, we use MaxConnections as the
> > + * threshold so we can prevent switching to the state unless we use
> > + * subtransactions.
> > + */
> > +#define MAX_HEAP_TXN_COUNT_THRESHOLD MaxConnections
> >
> > Isn't using max-heap equally effective in finding the largest
> > transaction whether there are top-level or top-level plus
> > subtransactions? This comment indicates it is only effective when
> > there are subtransactions.
>
> You're right. Updated the comment.
>
> I've attached the updated patches.
>

While reviewing the patches, I realized the comment of
binearyheap_allocate() should also be updated. So I've attached the
new patches.

Regards,

--
Masahiko Sawada
Amazon Web Services: https://aws.amazon.com