Re: Bug in batch tuplesort memory CLUSTER case (9.6 only)

Peter Geoghegan <pg(at)heroku(dot)com> writes:
> Attached patch adds a CLUSTER external sort test case to the
> regression tests, as discussed.

I took a quick look at this patch.

> This makes a hardly noticeable difference on the run time of the
> CLUSTER tests, at least for me. Consider the following:

I tried the patch on prairiedog's host. That's not the slowest
machine in the buildfarm, but it's right down there. The patch
increases the runtime of the "cluster" test from ~1 sec to ~3 sec,
which I agree is pretty negligible (total time for the standard
regression tests is ~5 min on this machine). That seems a cheap
price to pay for a significant improvement in code coverage.

What I don't much like is that it enlarges cluster.out with 200K of
random-looking, hard-to-manually-verify data. May I suggest that
we replace the SELECTs with

select * from
(select hundred, lag(hundred) over () as lhundred,
thousand, lag(thousand) over () as lthousand,
tenthous, lag(tenthous) over () as ltenthous from clstr_4) ss
where row(hundred, thousand, tenthous) <= row(lhundred, lthousand, ltenthous);
hundred | lhundred | thousand | lthousand | tenthous | ltenthous
---------+----------+----------+-----------+----------+-----------
(0 rows)

If you're good with that adjustment, I'm happy to commit this.

regards, tom lane