On Wed, Jun 18, 2014 at 4:48 AM, Stephen Frost <sfrost@snowman.net> wrote:
> I'm fine with having these start out as external tools which are doing
> checks, but I've been specifically asked about (and have desired myself
> from time-to-time...) an in-core capability to check index/heap/etc
> validity. Folks coming from certain other RDBMS's find it amazing that
> we don't have any support for that when what they really want is a
> background worker which is just automatically going around doing these
> checks.
Yes, I think that being able to verify the integrity of index and heap
relations is an important feature, and I think it's notable that we're
the only major RDBMS that lacks this support. I'm not quite sure what
that should entail just yet, but clearly it should be somewhat like
what I have here. I think the big open questions to make something
like this work are mostly around the cost/benefit ratio of each of the
checks I've outlined. Certainly, for that use case minimizing
disruption on a live system becomes even more important. I'll probably
look at a buffer access strategy, just to give an example of that off
the top of my head.
--
Peter Geoghegan