Re: Reducing some DDL Locks to ShareLock

On Tue, 2008-10-07 at 11:46 -0400, Tom Lane wrote:
> Simon Riggs <simon(at)2ndQuadrant(dot)com> writes:
> > On Tue, 2008-10-07 at 10:35 -0400, Tom Lane wrote:
> >> I wonder whether this could be helped if we refactored pg_constraint.
>
> > Sounds better. Doesn't make much sense as it is now.
>
> I looked at the code a bit, and it seems the only place where the
> current design makes any sense is in ChooseConstraintName, which
> explains itself thusly:
>
> * Select a nonconflicting name for a new constraint.
> *
> * The objective here is to choose a name that is unique within the
> * specified namespace. Postgres does not require this, but the SQL
> * spec does, and some apps depend on it. Therefore we avoid choosing
> * default names that so conflict.
> *
> * Note: it is theoretically possible to get a collision anyway, if someone
> * else chooses the same name concurrently. This is fairly unlikely to be
> * a problem in practice, especially if one is holding an exclusive lock on
> * the relation identified by name1.
>
> (The last bit of the comment falls flat when you consider constraints
> on domains...)
>
> Note that this policy is used for system-selected constraint names;
> it's not enforced against user-selected names. We do attempt (in
> ConstraintNameIsUsed) to reject duplicate user-selected constraint names
> *on the same object*, but that test is not bulletproof against
> concurrent additions. The refactoring I suggested would make for
> bulletproof enforcement via the unique indexes.
>
> To preserve the same behavior for system-selected constraint names with
> the new design, we'd still need to store namespace OIDs in the two new
> tables (I had been thinking those columns would go away), and still have
> nonunique indexes on (conname, connamespace), and probe both of the new
> catalogs via these indexes to look for a match to a proposed constraint
> name. So that's a bit of a PITA but certainly doable. Again, it's not
> bulletproof against concurrent insertions, but the existing code is not
> either.

How about we put a partial unique index on instead?

Dunno if its possible, but the above begins to sound too much froth for
such a small error.

--
Simon Riggs www.2ndQuadrant.com
PostgreSQL Training, Services and Support