On 2017-05-10 10:29:02 -0400, Tom Lane wrote:
> Petr Jelinek <petr.jelinek@2ndquadrant.com> writes:
> > On 10/05/17 07:09, Peter Eisentraut wrote:
> >> I think the correct fix is to have nextval() and ALTER SEQUENCE use
> >> sensible lock levels so that they block each other. Since
> >> nextval() currently uses AccessShareLock, the suggestion was for
> >> ALTER SEQUENCE to therefore use AccessExclusiveLock. But I think a
> >> better idea would be for nextval() to use RowExclusiveLock
> >> (analogous to UPDATE) and ALTER SEQUENCE to use
> >> ShareRowExclusiveLock, which would also satisfy issue #1.
>
> > When I proposed this upstream, Andres raised concern about performance
> > of nextval() if we do this, did you try to run any benchmark on this?
>
> As long as it doesn't block, the change in lock strength doesn't actually
> make any speed difference does it?
The issue isn't the strength, but that we currently have this weird
hackery around open_share_lock():
/** Open the sequence and acquire AccessShareLock if needed** If we haven't touched the sequence already in this
transaction,*we need to acquire AccessShareLock. We arrange for the lock to* be owned by the top transaction, so that
wedon't need to do it* more than once per xact.*/
This'd probably need to be removed, as we'd otherwise would get very
weird semantics around aborted subxacts. Upthread I theorized whether
that's actually still meaningful given fastpath locking and such, but I
guess we'll have to evaluate that.
Greetings,
Andres Freund
