On Sun, 2008-11-09 at 20:18 -0500, Tom Lane wrote:
> Simon Riggs <simon@2ndQuadrant.com> writes:
> > Will think some more and report back.
>
> If you want to do some more development, here's the portion of the
> patch as yet unapplied --- will save you extracting it for yourself.
Thanks.
More thought tells me that we should have a
LockRelationForCatalogUpdate() that uses nearly the same design pattern
as LockRelationForExtension(). There is no lockmode since we always take
the lock in exclusive mode.
Callers would grab the catalog update lock, re-read catalog, assemble
the new tuple, make in-place update and release lock. Lock is
non-transactional and exists only to serialise catalog updates from
concurrent DDL operations.
We then have the rule that all callers of heap_inplace_update() must
already hold the catalog update lock.
You like?
-- Simon Riggs www.2ndQuadrant.comPostgreSQL Training, Services and Support