>I have a question about the return code of the mpirun command.
>
>While there isn't anything on return code in the lam-6.5.9 man page
>for mpirun, in the mail archives there is a two-year old exchange
>on the suject "mpirun return code" which said:
>
> mpirun returns the return code from the first process in the
> job that died with a non-zero exit status. In the case that
> all processes died with a zero exit status, mpirun returns 0.
>
>Is this a supported feature of mpirun, i.e., can we count on this?
>Or did this feature not reach the supported status because of
>complications? For example, I would assume this behavior wouldn't
>hold if the -nw flag is used:^)
The return code of mpirun/lamexec is either an error-code for some
internal error in mpirun/lamexec, or the exit status of the first "child"
that exited abnormally, or 0 if all is well.
When used with the -nw flag, the only thing that changes is that the exit
status of the processes in the job have no bearing on mpirun's return
code. Hence, return code is either 0 if all is well, or non-zero if there
was an internal error in mpirun/lamexec.
This *is* a supported feature. This information was not on the
manpage only because of oversight. The manpages of mpirun and lamexec
in the CVS head and in the 7.0 release have now been updated to reflect
this feature.
Hope this helps.
--
Sriram Sankaran
email: ssankara_at_[hidden]
http://www.lam-mpi.org/
|
