Hi,
On Wed, Sep 24, 2025 at 08:31:27AM +0900, Michael Paquier wrote:
> We've had this exact same issue of a clock going backwards with one of
> the netbsd animals on an older version not supported anymore by
> upstream and that has been kicked out of the buildfarm, as far as I
> recall. This has created some disturbance in the regression tests
> causing EXPLAIN plan outputs we did not expect, in terms of extra
> negative signs and the animal showing red periodically.
This was the case initially on 32bit Hurd until I configured it to use
APIC (which is a requirement for HPET timers). So the clock is no longer
going backwards in an obvious way; and apart from the stats isolation
test, I have not seen issues in this area.
> So yes, this random factor would be annoying in the buildfarm.
How much timer resolution do we require from the system? GNU Mach seems
to (at least try to) guarantee that the timer won't go backwards, but it
does not guarantee (currently) that two consecutive clock_gettime()
calls will return something different in all cases.
Michael
