Re: Allow processes to reset procArrayGroupNext themselves instead of leader resetting for all the followers

On Sun, Nov 27, 2022 at 2:48 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
>
> Hi,
>
> On 2022-11-24 10:43:46 +0530, Bharath Rupireddy wrote:
> > While working on something else, I noticed that the proc array group
> > XID clearing leader resets procArrayGroupNext of all the followers
> > atomically along with procArrayGroupMember. ISTM that it's enough for
> > the followers to exit the wait loop and continue if the leader resets
> > just procArrayGroupMember, the followers can reset procArrayGroupNext
> > by themselves. This relieves the leader a bit, especially when there
> > are many followers, as it avoids a bunch of atomic writes and
> > pg_write_barrier() for the leader .
>
> I doubt this is a useful change - the leader already has to modify the
> relevant cacheline (for procArrayGroupMember). That makes it pretty much free
> to modify another field in the same cacheline.

Agreed. Thanks for the response.

--
Bharath Rupireddy
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com