There is an interesting option I_MPI_HYDRA_CLEANUP (and I_MPI_TMPDIR) documented, again for mpiexec.hydra, that supposed to create a file with list of PIDs; then, in case of problems, these can be fed to mpicleanup.
Somehow I've failed to get it working (using IMPI ver. 4.0.3.008). DOes this option work? Could somebody give me an example of its usage? Thank you very much!
-- Grigory Shamov HPC Analyst University of Manitoba
Keep in mind that the file created by enabling I_MPI_HYDRA_CLEANUP is deleted when the job ends, even if the job ends incorrectly. Its purpose is to provide a means of finding the jobs if Hydra ends incorrectly and the processes are still running. The file is named
PPID isthe parent process PID. I_MPI_TMPDIR is used to set the path to store this file. An example of the contents of the file: