On 16/01/2019 17:44, Robert Haas wrote:
> On Mon, Jan 14, 2019 at 1:41 PM Dimitri Fontaine <dimitri@citusdata.com> wrote:
>> One idea would be that if every temp table in the session belongs to the
>> transaction, and their namespace too (we could check through pg_depend
>> that the namespace doesn't contain anything else beside the
>> transaction's tables), then we could dispose of the temp schema and
>> on-commit-drop tables at PREPARE COMMIT time.
>
> Why not just drop any on-commit-drop tables at PREPARE TRANSACTION
> time and leave the schema alone? If there are any temp tables touched
> by the transaction which are not on-commit-drop then we'd have to
> fail, but as long as all the tables we've got are on-commit-drop then
> it seems fine to just nuke them at PREPARE time. Such tables must've
> been created in the current transaction, because otherwise the
> creating transaction aborted and they're gone for that reason, or it
> committed and they're gone because they're on-commit-drop. And
> regardless of whether the transaction we are preparing goes on to
> commit or abort, those tables will be gone afterwards for the same
> reasons. So there doesn't in this case seem to be any reason to keep
> them around until the transaction's fate is known.
Isn't that what happens already? PrepareTransaction() calls
PreCommit_on_commit_actions() from what I can tell.
--
Vik Fearing +33 6 46 75 15 36
http://2ndQuadrant.fr PostgreSQL : Expertise, Formation et Support
