Re: btree_gist valgrind warnings about uninitialized memory

On 2014-05-14 10:07:18 -0400, Tom Lane wrote:
> Heikki Linnakangas <hlinnakangas(at)vmware(dot)com> writes:
> > On 05/13/2014 05:13 PM, Andres Freund wrote:
> >> What's your plans with your spgist fix? Commit it once 9.5 is branched?
>
> > Good question. I don't know. I would still like to commit it to 9.4. It
> > doesn't require catalog changes, but it's an incompatible change in the
> > WAL record format. If we commit it to 9.4, it means that you cannot
> > replicate between 9.4beta1 and 9.4beta2. I think that's OK, but how do
> > others feel about that?
>
> I think that's an OK restriction as long as we warn people about it
> (you could update a replication pair as long as you shut them both
> down cleanly at the same time, right?). Can the WAL replay routine
> be made to detect incompatible records?

We could just bump the wal version. Somewhat surprisingly that works if
both nodes are shutdown cleanly (primary first)... But the errors about
it are really ugly (will moan about unusable checkpoints), so it's
probably not a good idea. Especially as it'll make it an issue for all
users, not just the ones creating spgist indexes.

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services