Andrew Dunstan <andrew@dunslane.net> writes:
> Tom Lane wrote:
>> Consistently? I'd have thought that problem would only fail once in a
>> while. It's hard to believe the timing would work out to make it a 100%
>> failure.
> You can see the history of the latest build runs here:
> http://www.pgbuildfarm.org/cgi-bin/show_history.pl?nm=spoonbill&br=HEAD
Remarkable. There is one run (2004-11-15) where it got past the rtree
test (and did indeed fail at seg) but the failure rate is certainly
upwards of 90%. Curious. There must be some effect that is
synchronizing the bgwriter's actions with the test sequence.
Back at the ranch, I am even more surprised to note that the bogus
seg output in the 11-15 run is different from what it is in today's.
There's not much I can do about it without access to a machine where
it's failing though. Can we get personal accounts on the buildfarm
machines?
regards, tom lane
