Actually, now that I look at it, I'm not sure what Jeff is talking about here is correct. I think Nathan's patch is in fact right.
Nathan's change doesn't in any way impact what gets passed to remote procs. All it does is modify what gets passed on the *orted's* command line. The orted has no idea what OMPI_foo means as an argument on its command line, and I don't think we want to change it. In orterun, we do correctly pickup any OMPI_xxx values and add them to the app's envars. This was not changed and will continue to be supported. So I think this patch is correct and okay as-is. On Oct 30, 2012, at 12:58 PM, Ralph Castain <r...@open-mpi.org> wrote: > I'll fix it, Jeff - the problem is that the plm/rsh was prepending a "-mca" > in those cases, when it shouldn't. Nathan's fix is close - I can fix the rest. > > > On Oct 30, 2012, at 12:52 PM, Jeff Squyres <jsquy...@cisco.com> wrote: > >> WAIT. >> >> This contradicts the intent of what I said on the call this morning. >> >> The point is that any env variable that begins with "OMPI_" is supposed to >> be propagated out to all the remote processes. It's a cheap/easy way for >> users to propagate their env variables to remote nodes (without needing to >> "mpirun -x" every variable they want to export). >> >> Specifically, I should be able to do something like this: >> >> ----- >> $ hostname >> my_localhost >> $ cat myscript >> : >> echo `hostname`: $OMPI_foo >> $ export OMPI_foo=bar >> $ cat hostfile >> some_remote_host >> $ mpirun --hostfile hostfile -np 1 myscript >> some_remote_host: bar >> $ >> ----- >> >> This behavior has been in OMPI for a long time; please do not take it out. >> >> If exporting non-MCA OMPI_<foo> env variables causes the problem, then it's >> a side effect. >> >> >> >> On Oct 30, 2012, at 3:40 PM, <svn-commit-mai...@open-mpi.org> wrote: >> >>> Author: hjelmn (Nathan Hjelm) >>> Date: 2012-10-30 15:40:04 EDT (Tue, 30 Oct 2012) >>> New Revision: 27526 >>> URL: https://svn.open-mpi.org/trac/ompi/changeset/27526 >>> >>> Log: >>> fix bug in plm/rsh that could add extraneous mca options to the orted argv >>> >>> cmr:v1.7 >>> >>> Text files modified: >>> trunk/orte/mca/plm/rsh/plm_rsh_module.c | 2 +- >>> >>> 1 files changed, 1 insertions(+), 1 deletions(-) >>> >>> Modified: trunk/orte/mca/plm/rsh/plm_rsh_module.c >>> ============================================================================== >>> --- trunk/orte/mca/plm/rsh/plm_rsh_module.c Tue Oct 30 15:23:15 2012 >>> (r27525) >>> +++ trunk/orte/mca/plm/rsh/plm_rsh_module.c 2012-10-30 15:40:04 EDT (Tue, >>> 30 Oct 2012) (r27526) >>> @@ -586,7 +586,7 @@ >>> * only if they aren't already present >>> */ >>> for (i = 0; NULL != environ[i]; ++i) { >>> - if (0 == strncmp("OMPI_", environ[i], 5)) { >>> + if (0 == strncmp("OMPI_MCA", environ[i], 8)) { >>> /* check for duplicate in app->env - this >>> * would have been placed there by the >>> * cmd line processor. By convention, we >>> _______________________________________________ >>> svn-full mailing list >>> svn-f...@open-mpi.org >>> http://www.open-mpi.org/mailman/listinfo.cgi/svn-full >> >> >> -- >> Jeff Squyres >> jsquy...@cisco.com >> For corporate legal information go to: >> http://www.cisco.com/web/about/doing_business/legal/cri/ >> >> >> _______________________________________________ >> devel mailing list >> de...@open-mpi.org >> http://www.open-mpi.org/mailman/listinfo.cgi/devel >
