Re: Patch to improve performance of replay of AccessExclusiveLock

On 22 March 2017 at 13:19, David Rowley <david(dot)rowley(at)2ndquadrant(dot)com> wrote:

>> Given that, do you agree to me applying assign_aels_against_subxids.v1.patch
>> as well?
>
> Does applying assign_aels_against_subxids.v1.patch still need to keep
> the loop to release the subxacts? Won't this be gone already with the
> subxact commit/abort record replays?

No it is still required because aborts and commits might have
subcommitted subxids.

> This does possibly mean that we perform more loops over the
> RecoveryLockList even if the subxact does not have an AELs, but its
> parent xact does. Wonder if this is a good price to pay for releasing
> the locks earlier?

We'd be performing the same number of loops as we do now. It's just
now they would have a purpose.

But we aren't doing it at all unless the top level xid has at least
one AEL, so the bulk of the problem is gone.

--
Simon Riggs http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services