On 2013-07-03 11:08:32 -0400, Tom Lane wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
> > On Wed, Jul 3, 2013 at 10:47 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> Are we somehow not going through ExecOpenIndices?
>
> > I dunno. I just did a quick black-box test:
>
> > CREATE TABLE foo (a int primary key);
> > BEGIN;
> > INSERT INTO foo VALUES (1);
> > SELECT relation::regclass, locktype, mode, granted FROM pg_locks;
>
> > I get:
>
> > relation | locktype | mode | granted
> > ----------+---------------+------------------+---------
> > pg_locks | relation | AccessShareLock | t
> > foo | relation | RowExclusiveLock | t
> > | virtualxid | ExclusiveLock | t
> > | transactionid | ExclusiveLock | t
>
> > No foo_pkey anywhere.
>
> That proves nothing, as we don't keep such locks after the query
> (and there's no reason to AFAICS). See ExecCloseIndices.
Should be easy enough to test by hacking LOCK TABLE to lock indexes and
taking out a conflicting lock (SHARE?) in a second transaction. I wonder
if we shouldn't just generally allow that, I remember relaxing that
check before (when playing with CREATE/DROP CONCURRENTLY afair).
Greetings,
Andres Freund
-- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training &
Services