On 2013-12-11 10:07:19 -0500, Tom Lane wrote:
> Andres Freund <andres@2ndquadrant.com> writes:
> > On 2013-12-10 19:55:12 -0500, Tom Lane wrote:
> >> We need a more consistent strategy for this :-(
>
> > Agreed, although I have no clue how it should look like. As a further
> > datapoint I'll add that installcheck already regularly fails in HEAD if
> > you have a HS standby connected via SR and hot_standby_feedback=on on
> > the standby. Some plans just change from index(only) scans to sequential
> > scans, presumably because of the lower xmin horizon changed the
> > stats. Since there's nothing running on the standby in those cases,
> > there has to be a pretty damn tiny window here somewhere.
>
> The case in create_index does a "vacuum analyze tenk1" and expects
> to get an index-only scan in the very next SQL command. So any delay
> in considering the table all-visible could break that test. I'm not
> sure if that's what you're talking about though. We could easily
> create some more delay for that case, for instance by moving the
> vacuum step to copy.sql as I was idly speculating about upthread.
> Do you remember offhand where the failures are?
No, but they are easy enough to reproduce. Out of 10 runs, I've attached
the one with the most failures and checked that it seems to contain all
the failures from other runs. All of them probably could be fixed by
moving things around, but I am not sure how maintainable that approach
is :/
Greetings,
Andres Freund
--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services