Further data: even when running via salloc ... bash the same thing
happens. Is there any way to tell SLURM to run in, for lack of a better
expression, "as if I were logged in via ssh" mode? That is to say emulate an
interactive run to a greater extent? If I issue the run command via ssh [run
command] [inputs] is works but anything run through SLURM just dumps the bash
script to a file(?)
Thanks,
~Mike C.
-----Original Message-----
From: Michael Colonno [mailto:[email protected]]
Sent: Wednesday, January 23, 2013 5:07 PM
To: slurm-dev
Subject: [slurm-dev] RE: not executing script(?)
Updating this thread: Iran additional experiments submitting the job
from the node it executes on - same behavior so I think this rules out system
config limits. It seems like the application runs scripts that run other
scripts and somehow SLURM's mode of execution confuses this. Anything else I
can test?
Thanks,
~Mike C.
-----Original Message-----
From: Moe Jette [mailto:[email protected]]
Sent: Tuesday, January 22, 2013 7:49 PM
To: slurm-dev; Michael Colonno
Subject: Re: [slurm-dev] not executing script(?)
Compare limits and environment variables for the two different modes of
operation.
Quoting Michael Colonno <[email protected]>:
>
> Hi ~
>
> Getting some odd behavior with SLURM I haven't seen before (2.5.0 on
> CentOS 6.3 x64 though I don't think any of that matters for this
> issue). I'm trying to run a code which launches from a bash script
> (commercial code, we didn't write it). If I ssh to a node and launch
> the code, everything works fine. Syntax looks like this:
>
> >> launch_script input_file
>
> If I paste the exact same command at the end on a srun command the
> job "runs" and I get a copy of the bash script that was supposed to
> have been executed in the directory I launched from (even with
> executable properties) in a file labeled input_file.[bunch of letters
> and numbers]. Syntax looks like:
>
> >>srun -n1 -p whatever launch_script input_file
>
> Scratching my head on this one. Clearly it finds the correct script
> to launch on the correct node but I can't explain the difference in
> behavior between the interactive and SLURM versions. Test cases like
> "hostname" all work fine. Probably not relevant but the parallel codes
> I've compiled into SLURM also launch and run great.
>
> Thanks,
> ~Mike C.
>
