"Heikki Linnakangas" <heikki@enterprisedb.com> writes:
> Tom Lane wrote:
>> Looking back, I think it was driven by the desire to tie the behavior
>> directly to things that are going to get persisted, such as locks.
>> From that standpoint your initial patch to attach a temp-check to
>> relation-drop 2PC entries would be the right kind of design. However,
>> given what we now know about the lock situation, I'd feel uncomfortable
>> with applying that without also fixing LockTagIsTemp, and right now
>> that's looking like much more complexity and possible performance
>> penalty than it's worth.
> Looking closer, this actually worked in 8.1, and was broken in 8.2 by
> this change:
Argh, so it's actually my bug :-(
> Before that, we had an isTempObject flag in LOCALLOCK, which worked even
> when the relation was dropped later on, unlike LockTagIsTemp.
Yeah. I guess my hindbrain was remembering that arrangement, because
adding such a field to LOCALLOCK was what I was first thinking about.
The problem though is that we need to take a lock on a table before
reading its pg_class row, in order to avoid race conditions when the
row is being deleted or updated. So we can't easily know at the time
the lock is taken whether it's a temp table or not. (In some contexts
such as the parser we might know which schema the table is in, but
most places are expected to be able to open and lock the table knowing
only its OID.) I think the only feasible solution like that would
involve calling the lock manager a second time to mark the lock temp
after we'd looked at the pg_class row. Which is not impossible, but
it would cost an extra hashtable search for each open of a temp table,
and it isn't really buying us anything compared to setting a global
flag in the same places.
> Anyway, patches attached, using the global flag approach, for 8.2 and
> 8.3. As discussed earlier, since the flag is global, we won't allow
> PREPARE TRANSACTION if you have operated on a temp table in an aborted
> subxact, but I think that's acceptable.
Patch looks good in a fast once-over, I'll check it more carefully
and apply today.
regards, tom lane