On Feb 9, 2010, at 5:34 PM, Jeff Squyres (jsquyres) wrote:
6. Could you send detail info about the issue (ini-file, mtt.log with verbose info and command line), we will look on that.
Let me reproduce and simplify; I was using a fairly complex ini file...
Oh, I see what happened -- I ran MTT manually on the command line and stepped through each of the phases manually, just to watch what was happening in each phase. Something like this
mtt --file foo.ini --verbose --mpi-get
mtt --file foo.ini --verbose --mpi-install
mtt --file foo.ini --verbose --test-get
mtt --file foo.ini --verbose --test-build
mtt --file foo.ini --verbose --test-run
If I run just a single mtt invocation, the submitting for mpi install and test build seems to work. However, there seems to be some built-in assumption that analyze::performance must be called...? I was just running the "trivial" suites and trying to submit that. Here's the --verbose output (I added the "GDS" verbose lines):
..lots of tests passing output
Test: cxx_ring, np=16, variant=8: Passed
Test: cxx_ring, np=16, variant=9: Passed
Test: cxx_ring, np=16, variant=10: Passed
Test: cxx_ring, np=16, variant=11: Passed
Test: cxx_ring, np=16, variant=12: Passed
### Test progress: 8 of 8 section test executables complete. Moving on.
Reporter MTTGDS: cached for later submit
*** Run test phase complete
*** Reporter finalizing
Submitted MPI Install to GDS
Submitted Test Build to GDS
*** WARNING: Could not load module MTT::Test::Analyze::Performance::: Can't
locate MTT/Test/Analyze/Performance/.pm in @INC (@INC contains:
/usr/lib/perl5/vendor_perl .) at (eval 1089) line 3.
..and oodles more errors just like this
Submitted Test Run to GDS
*** Reporter finalized
It looks like there is supposed to be some .pm file that I was supposed to have specified...? I'm not quite sure why, though -- performance analyzers shouldn't be necessary for all tests. We have oodles of correctness tests where performance isn't an issue, so that analysis is irrelevant.
One further question -- the initial email from Michael said that both libYAML and Syck are necessary. Why both? yaml.org says that Syck is the "old" interface and libYAML is preferred these days.
I'm testing bquery.pl -- unfortunately, I'm behind some proxies in the Cisco lab environment. I'll see if I can add proxy support...