Re: Compressing the AFTER TRIGGER queue

Robert Haas <robertmhaas(at)gmail(dot)com> writes:
> On Mon, Aug 1, 2011 at 1:42 PM, Dean Rasheed <dean(dot)a(dot)rasheed(at)gmail(dot)com> wrote:
>>> Don't we already do that when pruning HOT chains?

>> I thought that only happens after the transaction is committed, and
>> old enough, whereas the trigger code only needs to follow the chain in
>> the updating transaction.

> Hmm, true.

> I worry a bit that this might foreclose possible future optimization
> of the "self update" case, which is a known pain point. Am I wrong to
> worry?

I think it might be OK if you explicitly verify that xmin/cmin of the
linked-to tuple matches the (sub)transaction/command that queued the
trigger event. I don't recall whether the trigger code does that
already; I think there is some related test but it might not be that
strict.

There's also a definitional issue involved: if a transaction updates the
same tuple twice, in the presence of a deferred update trigger for the
table, is it supposed to (eventually) fire the trigger for both update
actions or only the last one? I have a feeling we might already be
locked into the second choice, but if not, this would probably force it.

regards, tom lane