Re: mvcc catalo gsnapshots and TopTransactionContext

On 2014-02-06 16:35:07 -0500, Tom Lane wrote:
> > I wonder though, if we couldn't just stop doing the
> > RelationReloadIndexInfo() for nailed indexes.
>
> No; you're confusing nailed indexes with mapped indexes. There are nailed
> indexes that aren't on mapped catalogs, see the load_critical_index calls
> in RelationCacheInitializePhase3.

Oh. I'd somehow remembered nailed catalogs would be a subset of mapped
ones. But as you say, they are not. Not sure I like that, but it's
certainly set for now.

> > Do you plan to backpatch this? If so, even to 8.4?

> I'm of two minds about that. I think this is definitely a bug, but we
> do not currently have any evidence that there's an observable problem
> in practice. On the other hand, we certainly get reports of
> irreproducible issues from time to time, so I don't care to rule out
> the theory that some of them might be caused by faulty cache reloads.
> That possibility has to be balanced against the risk of introducing
> new issues with this patch.
>
> Thoughts?

Let's let it stew a while in master, there certainly are enough
subtleties around this that I'd hesitate to add it to a point release in
the not too far away future. Doesn't seem that urgent to me. But after
that I'd say lets backpatch it.

Greetings,

Andres Freund

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