[torquedev] torque 4.1.2 login shells problem and bash -l workaround

Charles Henry chenry at ittc.ku.edu
Sun Oct 21 07:45:44 MDT 2012


Sorry to reiterate, pero:

Please, let me know if you've had this issue with torque 4.  
It would be helpful just to confirm that someone else shares this problem in a particular version of torque 4, or for someone else to tell me their installation works correctly.  
It's very important to me and I'm currently under a time crunch to get things working right.  After simplifying and streamlining the cluster for a wider audience, I fear having to introduce ugly workarounds to the cluster users.

Thanks,
Chuck


----- Original Message -----
> From: "Charles Henry" <chenry at ittc.ku.edu>
> To: torquedev at supercluster.org
> Sent: Thursday, October 18, 2012 1:34:57 PM
> Subject: torque 4.1.2 login shells problem and bash -l workaround
> 
> Hi list,
> 
> I have been following the torque 4 development, and I'm currently
> using torque 4.1.2 on RHEL6.2.  I have found that I cannot get
> cluster jobs to run correctly without using "#!/bin/bash -l" in each
> script.  A few sites (academic and government) are listing this
> workaround in their cluster FAQs.
> 
> Our site uses mpi-selector and needs to source /etc/profile for every
> cluster job (interactive or not).  I'm going to get a million "why
> is mpiexec not found questions" if I have to rely on the workaround
> instead of addressing the problem.  I have looked for settings in
> the documentation and read the source code.
> 
> The relevant settings are defined globally inside
> src/resmom/mom_main.c
> ... (line 205)
> int      src_login_batch = TRUE;
> int      src_login_interactive = TRUE;
> ...
> 
> and used in src/resmom/start_exec.c
> ... (line 3736)
> void source_login_shells_or_not(
> ...
>   if (((TJE->is_interactive == TRUE) && (src_login_interactive ==
>   FALSE)) ||
>       ((TJE->is_interactive != TRUE) && (src_login_batch == FALSE)))
> ...
> 
> Where those values are declared as "extern int", so the values from
> mom_main.c are accessible once the binaries are linked.
> 
> There's no error message from the source_login_shells_or_not
> function, and the code looks very similar to the torque-3 code
> (except for being wrapped up into functions).  Can anyone shed some
> light on the problem?
> 
> Chuck


More information about the torquedev mailing list