|
|
Boost Testing : |
Subject: Re: [Boost-testing] [EXTERNAL] Re: Suddenly takes > 9hours to test?
From: Belcourt, K. Noel (kbelco_at_[hidden])
Date: 2012-02-13 11:37:08
Hi Steve,
On Feb 12, 2012, at 9:19 PM, Steve M. Robbins wrote:
> On Mon, Feb 06, 2012 at 05:56:23PM +0000, Belcourt, Kenneth wrote:
>>
>> I neglected to mention that I made a change to the trunk build tools
>> in rev. 76862. I tested this change set with the Sandia testers for
>> a month so I felt pretty good about it but I'm now concerned I may
>> have broken FreeBSD or other unix testers. This change shouldn't be
>> in release, only in trunk.
>
> The reason it affects release is that the release testing scripts uses
> the TRUNK version of boost build.
Yes, I'd forgotten about that, thanks for reminding me.
> Can you try building with an earlier
>> trunk rev. and see if the behavior improves?
>
> I have just now verified that rev 76861 is OK and 76862 behaves badly.
Good, at least we know what's causing the problem.
> I'm happy to revert the change if it's causing any problems, just
> let me know.
>
> I think that's best.
I'll attempt to get this reverted today. Note that there's a work
around using the patch_boost script that might also work for you. You
could take working copies of jam.h, jam.c and execunix.c and copy them
over the committed versions.
> I'm trying to figure out the issue. As best as I understand at
> present, the "mpi-run" lines in bjam result in the command line
> "mpirun -np 2 distributed_property_map_test" (etc). If run by hand,
> this command finishes normally. When run by bjam, the two spawned
> distributed_property_map_test processes seem to execute properly
> (output is written to appropriate file) but they don't finish -- the
> processes hang around as zombies. The mpirun process seems to beSo
> stuck in a poll() loop, according to strace.
> However, there is also a bunch of child process handling stuff. Is
> that related to the "-mN" option in some way?
Yes, handling zombies and handling output are tied, that's why I
committed them together. Are you using shared memory MPI
implementations?
Should be reverted with commit rev. 77005. Note that with this commit
reverted, we'll have some new Sandia test failures and several Sandia
testers will fail to cycle daily. Steve, we'll have to work on
figuring out what I've broken on your platform. Perhaps we could talk
offline.
-- Noel