On Tue, 2006-06-06 at 16:11, Tom Lane wrote:
> Scott Marlowe <smarlowe@g2switchworks.com> writes:
> > The cost of not aborting on the first sort is that you may never see
> > what the part of the plan is that's killing your query, since you never
> > get the actual plan.
>
> Well, you can get the plan without waiting a long time; that's what
> plain EXPLAIN is for. But I still disagree with the premise that you
> can extrapolate anything very useful from an unfinished EXPLAIN ANALYZE
> run. As an example, if the plan involves setup steps such as sorting or
> loading a hashtable, cancelling after a minute might make it look like
> the setup step is the big problem, distracting you from the possibility
> that the *rest* of the plan would take weeks to run if you ever got to
> it.
Sure, but it would be nice to see it report the partial work.
i.e. I got to using a nested loop, thought there would be 20 rows,
processed 250,000 or so, timed out at 10 minutes, and gave up.
I would find that useful.