On Mon, Apr 13, 2009 at 7:06 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> Josh Berkus <josh@agliodbs.com> writes:
>> I would suggest putting this info in a separate table, pg_change. It
>> would have oid, catalog, user_changed, changed_on. That way we could
>> simply keep the data for all objects which have an OID.
>
> That makes more sense to me --- it would easily extend to all cases
> and would not impose any overhead (in the form of useless columns)
> for catalogs that you didn't want to track in a particular case.
>
> The main problem that would have to be considered is how to flush
> no-longer-useful entries (which of course entails deciding which
> those are).
I kinda think that the only thing that's going to make sense here is
to drop the pg_change entries when the object is dropped. Now,
admittedly, that means you can't track drops. But otherwise, you have
the potential for pg_change to get really big and full of cruft, and I
don't think there's going to be an easy way to garbage collect it.
I really like the basic design, though.
...Robert