Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ptp-user] Debug on a SLURM cluster through PBS resource manager

Looking forward to seeing next improvements then :-) Great job by the way!

I used a submission script before...  How can I use the SLURM rm to submit my script? Change "Application program" of Application tab in Run Configurations? What kind of command then? The slurm proxy is still a black box for me...

Is using the submission script still provide me infos in the jobs list or anywhere else?

Thanks!

Steph 

On 25 oct. 2011, at 17:41, Greg Watson wrote:

> Stephanie,
> 
> The SLURM resource manager hasn't been transitioned to the new JAXB framework yet, so doesn't have the same flexibility.
> 
> For your first question, I'd suggest using a script on the target system that does the data copy, then runs the job. You can then use the SLURM RM to submit the script rather than the executable.
> 
> For the second question, unless the SLURM launch configuration already provides the ability to change these properties then unfortunately you can't. Hopefully when SLURM is transitioned to the new framework you'll be able to change these.
> 
> Regards,
> Greg
> 
> On Oct 25, 2011, at 4:36 AM, Stéphane Backaert wrote:
> 
>> Thanks for you reply.  I made some progress: I can launch a job with the SLURM resource manager. I see my job in the jobs list :-))
>> 
>> But the cluster which I currently use has a scratch space implemented as one disk per node, no real parallel FS layer. So I need to copy my data on each node disk and move them back at the end. I usually do this pre and post process in my submission script (like srun cp $HOME:data node:/scratch/)...
>> 
>> -my question: how can I set up the resource manager to use my submission script?
>> -and/or my second question: how can I change some properties like job name, std output file name... ? Because I do not see any output... Thanks to the JAXB stuff?
>> 
>> Best regards,
>> Stephane 
>> 
>> 
>> On 19 oct. 2011, at 15:38, Greg Watson wrote:
>> 
>>> 
>>> On Oct 19, 2011, at 3:21 AM, Stéphane Backaert wrote:
>>> 
>>>> Hello,
>>>> 
>>>> I have access to a cluster via ssh. This cluster uses the SLURM job manager. I can connect both a SLURM resource manager or a PBS-Generic-Batch rm. 
>>>> The former shows me the icons relative to the nodes in the "Machines" view but the submitted jobs via this interface seem to be launch on the frontend and complain about missing libs. Is SLURM rm an interactive launcher?
>>>> The latter does not show me any information about the cluster (nothing in Machines or Jobs List views) but the jobs start correctly (SLURM understands PBS keyword in the submission script?).
>>> 
>>> SLURM is not interactive as far as I know. It may be an issue with the version of SLURM you have installed. Jie, can you help?
>>> 
>>>> 
>>>> My problem: when I try to launch a parallel debug job with the PBS-Generic-Batch rm, it complains with the message "Configuration is missing run commands for job submission 778c1a5f-d3db-41ce-8be3-f3736e47b7a1 debug". Any clue?
>>> 
>>> 
>>> Do you have PBS installed on the cluster also? I don't believe SLURM is supported by the new resource manager framework yet.
>>> 
>>> Regards,
>>> Greg
>>> 
>>>> 
>>> 
>>> _______________________________________________
>>> ptp-user mailing list
>>> ptp-user@xxxxxxxxxxx
>>> https://dev.eclipse.org/mailman/listinfo/ptp-user
>> 
>> _______________________________________________
>> ptp-user mailing list
>> ptp-user@xxxxxxxxxxx
>> https://dev.eclipse.org/mailman/listinfo/ptp-user
> 
> _______________________________________________
> ptp-user mailing list
> ptp-user@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/ptp-user



Back to the top