At Thursday 24 June 2010, Ralf Wildenhues wrote:
> * Stefano Lattarini wrote on Wed, Jun 23, 2010 at 03:04:07PM CEST:
> > May I ping this? If you think the patch is OK, I'd like to see
> > it applied, since at present I'm continuing to experience
> > annoying spurious failures in cond5.test.
>
> Thanks for the reminder. Yes, the patch looks good for maint if
> you see spurious failures.
Yes, I do whenever I run many checks at the same time with lower
priority (e.g. "nice -n19 make -j32 check"), which I do quite often.
> Do you have an old system?
Well, it's not new :-)
$ cat /proc/cpuinfo
processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 6
model name : AMD Athlon(tm) XP 1800+
stepping : 2
cpu MHz : 1532.918
cache size : 256 KB
...
$ free | awk '(NR == 2) { print $2 }'
774904
> What was
> the highest $try that you needed, 30 seems a bit excessive, no?
I preferred to err on the side of caution. After all, if the test
script works correctly, it exits much earlier than after 30 tries
(usually 1 try is enough). Also, I'm not expecting to see the bug it
looks for cropping up often, so even if the test takes 5 minutes in
the unlikely situation of a bug's reapperence, that's not a problem
IMHO.
> I wonder what the current lower bound on PID reuse is on systems.
> There are certainly systems which use only 32K PIDs, and process
> creation can easily be thousand per second. I hope that 10
> seconds are still safe.
I think that the possibility of a spurious failure here is very very
low. But I might be wrong, and it would fine by me having, say, 100
tries every 3 seconds instead of 30 tries every 10 seconds. Your
call.
Thanks,
Stefano
