MPI falsely detects SGE grid environemnt mode

[AhmeddHanyy]

I enter a machine with qrsh -pe <parallel_environment_name> 30
When I run in the terminal mpiexec -n 1 sleep 5, I get the following error:

error: executing task of job 37467491 failed: failed sending task to [email protected]: can't connect to service
error: commlib error: got select error (poll() or select() error: Connection refused)
[mpiexec@dvtpe09] ui_cmd_cb (/bata/ahmabd4u/mpich-4.2.2-ch3/mpich-4.2.2/src/pm/hydra/mpiexec/pmiserv_pmci.c:51): Launch proxy failed.
[mpiexec@dvtpe09] HYDT_dmxu_poll_wait_for_event (/bata/ahmabd4u/mpich-4.2.2-ch3/mpich-4.2.2/src/pm/hydra/lib/tools/demux/demux_poll.c:76): callback returned error status
[mpiexec@dvtpe09] HYD_pmci_wait_for_completion (/bata/ahmabd4u/mpich-4.2.2-ch3/mpich-4.2.2/src/pm/hydra/mpiexec/pmiserv_pmci.c:173): error waiting for event
[mpiexec@dvtpe09] main (/bata/ahmabd4u/mpich-4.2.2-ch3/mpich-4.2.2/src/pm/hydra/mpiexec/mpiexec.c:260): process manager error waiting for completion

When I unset the environment variable PE_HOSTFILE, it works fine.

Do I need to always unset the environment variable to be able to run ?

[raffenet]

Unsetting the environment variable is fine as a workaround, but if the error is consistent our integration may need updating. Can you create an issue? Replicating an SGE environment may be challenging in order to fix. Could you also run mpiexec with -v to get more verbose output?