Thread: Operational note: possible infinite loop in today's HEAD testing
If you run any PG buildfarm critters that are configured with
--enable-tap-tests, you might want to check to see if they're caught
in an infinite loop in pgbench :-(. If so, just kill -9 that process,
and the next run should recover OK.
The cause is a platform-dependent bug introduced in commit 12788ae49
and fixed in 9779bda86.
A look through the buildfarm status says that the only members using
--enable-tap-tests that haven't reported in successfully despite the bug
are hamster, jacana, and topminnow. But I'm sending this out more
widely just in case I missed somebody's machine.
(Andrew: I wonder if we should have a sanity limit on how long buildfarm
runs are allowed to take. Not sure what to put the limit at, though.)
regards, tom lane
On 09/26/2016 09:16 PM, Tom Lane wrote: > If you run any PG buildfarm critters that are configured with > --enable-tap-tests, you might want to check to see if they're caught > in an infinite loop in pgbench :-(. If so, just kill -9 that process, > and the next run should recover OK. > > The cause is a platform-dependent bug introduced in commit 12788ae49 > and fixed in 9779bda86. > > A look through the buildfarm status says that the only members using > --enable-tap-tests that haven't reported in successfully despite the bug > are hamster, jacana, and topminnow. But I'm sending this out more > widely just in case I missed somebody's machine. > > (Andrew: I wonder if we should have a sanity limit on how long buildfarm > runs are allowed to take. Not sure what to put the limit at, though.) Maybe, it would have to be configurable. We actually have some code for doing that sort of thing from back in the CVS days when we used to get CVS hangs, so it shouldn't be hard. cheers andrew
On 09/26/2016 09:16 PM, Tom Lane wrote: > If you run any PG buildfarm critters that are configured with > --enable-tap-tests, you might want to check to see if they're caught > in an infinite loop in pgbench :-(. If so, just kill -9 that process, > and the next run should recover OK. > > The cause is a platform-dependent bug introduced in commit 12788ae49 > and fixed in 9779bda86. > > A look through the buildfarm status says that the only members using > --enable-tap-tests that haven't reported in successfully despite the bug > are hamster, jacana, and topminnow. But I'm sending this out more > widely just in case I missed somebody's machine. > > (Andrew: I wonder if we should have a sanity limit on how long buildfarm > runs are allowed to take. Not sure what to put the limit at, though.) Maybe, it would have to be configurable. We actually have some code for doing that sort of thing from back in the CVS days when we used to get CVS hangs, so it shouldn't be hard. cheers andrew