Re: [Boost-testing] Result statuses for run failures of MPI tests

Subject: Re: [Boost-testing] Result statuses for run failures of MPI tests
From: KTC (ktc_at_[hidden])
Date: 2010-07-05 22:38:11

On 04/07/2010 21:39, Jeremiah Willcock wrote:
>
> This issue is still appearing, in particular on Visual C++. Would it be
> possible for someone to look at the logs for that platform and see if
> the tests are actually failing? Also, are the raw test logs available
> anywhere? I might be able to look through them myself to see if there is
> a real bug in BGL or not.
>

I'm guessing that's me. :)

Yes it is actually failing during runtime. Like I posted last time in 
response to this, if one look at bjam.log something along the following 
lines can be seen for adjlist_build_test-2:

====== BEGIN OUTPUT ======
n = 10000, p = 0.003, seed = 1278372241
Building graph with the iterator constructor... done.
Building graph with add_edge from the root...synchronizing...done.
Verifying graphs...done.
Building graph with add_edge from everywhere...
job aborted:
rank: node: exit code: message
0: ktc-dt-brother: -1: aborting MPI_COMM_WORLD, error -1, comm rank 0
1: ktc-dt-brother: terminated

---- error analysis -----

0: 
C:\src\bgl-test\results\boost\bin.v2\libs\graph_parallel\test\adjlist_build_test-2.test\msvc-10.0\debug\link-static\threading-multi\adjlist_build_test-2.exe 
aborted the job on ktc-dt-brother. abort code -1

---- error analysis -----

EXIT STATUS: -1
====== END OUTPUT ======
...failed capture-output

Well, okay, looking at the chart again, some not all of them are failing 
during runtime. There's some compile time and link time fail that can be 
seen as well.

distributed_connected_components_test-2 is exceeding (the default) time 
limit during run. That one might eventually succeed if left long enough 
I'm not sure, as it's ususally only nonblocking_test-1 from MPI that's 
still hanging around at the end of a complete test.

distributed_strong_components_test-2 I have runtime timeout on one, and 
runtime fail on another.

Ignore the fail for distributed_st_connected_test-2 & 
hohberg_biconnected_components_test-2 in KTC-VC10_BGL. I think that's 
just the computer not coping and hence not compiling before the time 
limit. :S

mesh_generator_test-2 I have a pass on one run and a fail in another...

I was going to post the raw log, but I'm going to run the tests again on 
its own with a extra long timeout value and see what happens first.

KTC