Andrew Dunstan <andrew.dunstan@2ndquadrant.com> writes:
> On 07/25/2017 11:25 AM, Tom Lane wrote:
>> Oh. So when was the last time it worked? And why do the buildfarm
>> archives contain apparently-successful log_stage entries for pg_ctl-check
>> on jacana, up through yesterday when I looked?
> There was a buildfarm bug, since corrected, which was not properly
> picking up errors in a corner case. You will see the errors if you look
> at all the logs for pg_ctl-check from 12 days back if they exist. The
> last known actual good run of this test was 33 days ago, i.e. 2017-06-22
Hm. This failure presumably started with commit f13ea95f9, which
introduced use of command_like() in the pg_ctl TAP test; but that
didn't go in until 2017-06-28. Was jacana not running in the
interim?
Anyway, if we believe that it broke with f13ea95f9, hen it's plausible
that CMD.EXE has been sharing pg_ctl's stdout/stderr all along, and we
just had not noticed before. Could you check what happens if we
change the bInheritHandles parameter as I suggested upthread?
regards, tom lane