Skip to main content

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [ptp-user] Correction ---- ptp/slurm: How to redefine location of .eclipsesettings -- $HOME is not mounted on compute nodes.

I would start by uncommenting line 285 of slurm-generic-batch.xml. I’m not sure what the default file-staging-location should be, but I think this should be the same as the “directory” attribute of the submit-batch command.

If this doesn’t help, you can start trying to debug this by enabling the debugging options in Preferences>Parallel Tools>Target Configurations. In particular, the Command Debugging Options should show what command is actually being run. You could also monitor the .eclipsesettings directory to make sure the file is getting created (it may have a different name each launch, I’m not sure).

If this proved unhelpful, I would move to debugging the source. Let me know if you would like more information on this.

Greg

On Mar 7, 2016, at 1:04 PM, John Robinson <jsrobin@xxxxxxxxx> wrote:





It appears that the error occurs even in cases where the home directory is mounted on compute nodes.

I submitted the the problem, along with a similar problem for debug launches as
bug id: 489103 . After submitting the bug, I found that a similar issue had occured last year as bug id: 479083 .

Do you have any idea as to how to best "debug" this problem?


Sincerely,
John Robinson




--------------------------------------------------------------------------------------------------------

Date: Tue, 1 Mar 2016 13:03:28 -0500
From: Greg Watson <g.watson@xxxxxxxxxxxx>
To: PTP User list <ptp-user@xxxxxxxxxxx>
Subject: Re: [ptp-user] ptp/slurm: How to redefine location of
        .eclipsesettings -- $HOME is not mounted on compute nodes.
Message-ID: <AD36A169-CA18-494D-9790-9B928F9224B4@xxxxxxxxxxxx>
Content-Type: text/plain; charset=utf-8

I?m a bit puzzled as to why it matters if the home directories are mounted on the compute nodes or not. Normally the batch script would be placed in $HOME/.eclipsettings/?managed_file_for_script, then the sbatch command executed with this path as an argument. This would happen on the login node, so should be independent of the compute nodes. Can you try creating a simple script, put it in $HOME/.eclipsesettings/foo, then run ?sbatch $HOME/.eclipsesettings/foo? from the command line on the login node?

Greg

> On Mar 1, 2016, at 9:59 AM, John Robinson <jsrobin@xxxxxxxxx> wrote:
>
>
> The subject spells out the root of the problem.
>
> When PTP (on generic slurm) prepares a run configuration, it puts the
> related batch file in $HOME/.eclipsesettings directory.
>
> If the aforementioned directory is not on the compute nodes, then the launch will fail
> with an error ---- "sbatch Exited with value: 1, sbatch err: Unable to open file .eclipsesettings/ccc....managed_file_for_script  ..... job submit failed"
>
>
> Changing the value of the environment variable HOME in the ".bashrc" or in the ptp configuration information dialog used to configure remote module loading will not change the outcome (the code may be using "getent passwd" which will ignore $HOME settings).
>
> The "home" directory is used to specify ptpDirectory and other directories and files which may not exist on the compute nodes.
>
> This means in particular, on large numbers of compute clusters and HPCs ( I have never worked on one in which "home" directories are mounted ) the Eclipse/PTP project is horrible broken "out of the box".
>
> Is there some fix that will work in this case (home directory never mounted on compute nodes) ?
>
> _______________________________________________
> ptp-user mailing list
> ptp-user@xxxxxxxxxxx
> To change your delivery options, retrieve your password, or unsubscribe from this list, visit
> https://dev.eclipse.org/mailman/listinfo/ptp-user







Back to the top