On Mon, Jul 1, 2024 at 2:18 AM Noah Misch <noah@leadboat.com> wrote:
> On Tue, Mar 12, 2024 at 02:10:59PM +0200, Alexander Korotkov wrote:
> > I'm going to push this if no objections.
>
> Commit e85662d wrote:
> > --- a/src/backend/storage/ipc/procarray.c
> > +++ b/src/backend/storage/ipc/procarray.c
>
> > @@ -2740,6 +2741,8 @@ GetRunningTransactionData(void)
> > */
> > for (index = 0; index < arrayP->numProcs; index++)
> > {
> > + int pgprocno = arrayP->pgprocnos[index];
> > + PGPROC *proc = &allProcs[pgprocno];
> > TransactionId xid;
> >
> > /* Fetch xid just once - see GetNewTransactionId */
> > @@ -2760,6 +2763,13 @@ GetRunningTransactionData(void)
> > if (TransactionIdPrecedes(xid, oldestRunningXid))
> > oldestRunningXid = xid;
> >
> > + /*
> > + * Also, update the oldest running xid within the current database.
> > + */
> > + if (proc->databaseId == MyDatabaseId &&
> > + TransactionIdPrecedes(xid, oldestRunningXid))
> > + oldestDatabaseRunningXid = xid;
>
> Shouldn't that be s/oldestRunningXid/oldestDatabaseRunningXid/?
Thank you for catching this.
> While this isn't a hot path, I likely would test TransactionIdPrecedes()
> before fetching pgprocno and PGPROC, to reduce wasted cache misses.
And thanks for suggestion.
The patchset is attached. 0001 implements
s/oldestRunningXid/oldestDatabaseRunningXid/. 0002 implements cache
misses optimization.
If no objection, I'll backpatch 0001 and apply 0002 to the head.
------
Regards,
Alexander Korotkov
Supabase