On Wed, Jan 02, 2019 at 05:18:08PM -0500, Tom Lane wrote:
> Hm. I can reproduce this if I start in a virgin database, but not
> otherwise. I think the problem is that the prepared transaction has
> created the pg_temp_NN schema for its session, and therefore there
> is an uncommitted pg_namespace entry for that schema name, and the
> second session is also trying to create that schema (because it has
> the same backend ID) so it blocks waiting to see if that index
> entry commits.
That should not be allowed to commit directly. I think that we should
just add a new value for MyXactFlags which tracks the transaction
where the temporary namespace has been created, and generate an error
if trying to use 2PC in this case. A separate flag looks necessary to
me so as we don't bump on "cannot PREPARE a transaction that has
operated on temporary tables" in this case. That would be
back-patchable, and if both XACT_FLAGS_ACCESSEDTEMPREL and the new
flag are set then we just blow up on the original error message when
committing.
Opinions?
--
Michael