On Mon, Jan 19, 2015 at 12:05:01PM -0500, Tom Lane wrote:
> Robert Haas <robertmhaas@gmail.com> writes:
> > On Mon, Jan 19, 2015 at 11:30 AM, Andres Freund <andres@2ndquadrant.com> wrote:
> >> Sure, but the log isn't invisible. As mentioned one paragraph above, I
> >> don't think it's likely to ever be noticed in the client code in the
> >> cases where it happens in production.
>
> > Yes, that is my feeling as well.
>
> Meh. I still don't like it, but I guess I'm outvoted. Unless there are
> further votes, what we have at this point is just:
>
> - elog(WARNING, "pgstat wait timeout");
> + ereport(LOG, (errmsg("using stale statistics instead of current ones because stats collector is not
responding")));
>
> with no conditionality for pgstat_vacuum_stat vs. other callers.
That is satisfactory to me, too.