[ET Trac] [Einstein Toolkit] #1075: Test cases with specific number of processes

[Einstein Toolkit]

#1075: Test cases with specific number of processes
--------------------------+-------------------------------------------------
  Reporter:  eschnett     |       Owner:            
      Type:  enhancement  |      Status:  reopened  
  Priority:  major        |   Milestone:  ET_2013_11
 Component:  Cactus       |     Version:            
Resolution:               |    Keywords:            
--------------------------+-------------------------------------------------

Comment (by hinder):

 I believe that this can be made to work with SimFactory, but it will
 require modifications to SimFactory.  When SimFactory needs to run the
 tests as part of a simulation, it does this by using the file
 simfactory/bin/RunTestSuite as its "run script", passing the real machine
 runscript as an argument to that script.  RunTestSuite then sets
 CCTK_TESTSUITE_RUN_COMMAND to call the passed runscript on '$parfile',
 which is supplied by the test system.  The test system will also supply
 '$nprocs', since this command is usually an mpirun command, but simfactory
 ignores this at present.  The reason is because the runscript has already
 been expanded at this point, so that @NUM_PROCS@ has been replaced with
 the number of processes that the whole simulation is running on.  One
 solution would be to modify simfactory so that, in the case of running a
 testsuite, it expands @NUM_PROCS@ to a reference to an environment
 variable, e.g. $NUM_PROCS.  The CCTK_TESTSUITE_RUN_COMMAND set in
 RunTestSuite can then set this variable to '$nprocs' before launching the
 machine's run script.  This is similar to what is done for the parameter
 file; there is special code in simfactory to look at a TESTSUITE_PARFILE
 environment variable which is passed by the CCTK_TESTSUITE_RUN_COMMAND.

 This assumes that changing only NUM_PROCS in the machine run script is
 sufficient to do what we want; other variables might have been used, and
 some mpirun commands might take the number of cores or something similar
 instead.  I expect this to break some systems.

-- 
Ticket URL: <https://trac.einsteintoolkit.org/ticket/1075#comment:37>
Einstein Toolkit <http://einsteintoolkit.org>
The Einstein Toolkit