Closed christopherwharrop-noaa closed 8 months ago
A couple questions here:
parsl_flux_mpi_pi1_run.out
or parsl_flux_mpi_pi2_run.out
. Are these files automatically removed? I see in test_parsl_flux_mpi_hello.py
in test_run_mpi_pi
you are testing them against each other, but don't see them in the repository. I'm on the main
branch. Is there a way to get to these files and see the contents?HH:MM:SS
. Could you point me to the Fprtran code that needs to be updated? I think an appropriate place would be in mpi_pi.f90
.My idea would be to start with the timestamp first, ensure that is outputting the correct timestamp with the date included in each of parsl_flux_mpi_pi1_run.out
and parsl_flux_mpi_pi2_run.out
. Then continue on to the confirmation that the tests run concurrently.
The output files parsl_flux_mpi_pi1_run.out
and parsl_flux_mpi_pi2_run.out
should be produced when you run the tests.
The code that needs updating is mpi_pi.f90
.
I agree. Fix the time stamp first because you have to have that before you can do anything else.
The
test_run_mpi_pi
test intest_parsl_flux_mpi_hello.py
ensures that two runs of thempi_pi.exe
program do not share nodes. However, the test does not actually verify that the tests run concurrently.The output of the tasks is written in
parsl_flux_mpi_pi1_run.out
andparsl_flux_mpi_pi2_run.out
. This output includes time stamps for the start and end times of the run. In order to prove that the two tests run concurrently, two additional assertions are needed. We need to guarantee that the start time ofpi1
is not greater than the end time ofpi2
and that the start time ofpi2
is not greater than the end time ofpi1
.An additional problem is that the time stamps only include
HH:MM:SS
. If the execution of the code spans midnight, then the validation of concurrent execution times will not work. The Fortran code needs to be updated to write out the full time stamp such that it includes the date as well as the time.