Output redirection to files broken in development branches

[albandil]

I am using self-compiled development version of Open MPI on several version of openSUSE. The output redirection command line options --output-filename and --output-directory of mpiexec are broken in the current master branch as well as in the v5.0.x branch. No files or directories are created when using the commands

mpiexec --output-filename log hostname
mpiexec --output-directory logs hostname

Here follow some results from my investigation via git bisect:

master branch

• Pull request Update IOF redirection options #7162 introduced options --output-filename and --output-directory (both working well)
• Pull request Remove ORTE project #7202 has broken --output-directory (output directory no longer produced)
• Pull request Pass oversubscribe status to MPI layer #8998 has broken --output-filename (output files no longer produced)

v5.0.x branch

• (Pull request Update IOF redirection options #7162 common with master)
• (Pull request Remove ORTE project #7202 common with master)
• Pull request v5.0.x: Pass oversubscribe status to MPI layer #9026 has broken --output-filename (output files no longer produced)

I consider the per-rank output redirection a great added value of Open MPI compared to other implementations and would welcome if this functionality was fixed.

I am aware that there is also some additional option --output, but at least in v5.0.x (haven't tried master) I am not able to reproduce --output-filename with that switch either.

[rhc54]

It actually appears to be working fine in master:

$ mpirun --output file=log hostname
$ ls log*
log.prterun-rhc-node01-22901@1.01.err  log.prterun-rhc-node01-22901@1.06.err  log.prterun-rhc-node01-22901@1.12.out
log.prterun-rhc-node01-22901@1.01.out  log.prterun-rhc-node01-22901@1.06.out  log.prterun-rhc-node01-22901@1.13.out
log.prterun-rhc-node01-22901@1.02.err  log.prterun-rhc-node01-22901@1.07.err  log.prterun-rhc-node01-22901@1.14.out
log.prterun-rhc-node01-22901@1.02.out  log.prterun-rhc-node01-22901@1.07.out  log.prterun-rhc-node01-22901@1.15.out
log.prterun-rhc-node01-22901@1.03.err  log.prterun-rhc-node01-22901@1.08.err  log.prterun-rhc-node01-22901@1.16.out
log.prterun-rhc-node01-22901@1.03.out  log.prterun-rhc-node01-22901@1.08.out  log.prterun-rhc-node01-22901@1.17.out
log.prterun-rhc-node01-22901@1.04.err  log.prterun-rhc-node01-22901@1.09.err  log.prterun-rhc-node01-22901@1.18.out
log.prterun-rhc-node01-22901@1.04.out  log.prterun-rhc-node01-22901@1.09.out  log.prterun-rhc-node01-22901@1.19.out
log.prterun-rhc-node01-22901@1.05.err  log.prterun-rhc-node01-22901@1.10.out
log.prterun-rhc-node01-22901@1.05.out  log.prterun-rhc-node01-22901@1.11.out

I noticed that the output for the local procs isn't getting correctly stored, so I'll look into that bug. But the cmd line option appears to be working.

That said, you should have gotten a deprecation warning for the old options. I'll have to fix that as well.

[albandil]

I can confirm that --output file=... works well. I must have been using wrong keyword ("filename"?).

[rhc54]

filename should also have worked - we only check the "file" portion of the string.

[rhc54]

I'll update OMPI once the PRRTE fix has been committed.