Parpool errors on SLURM computing system

I'm running a script that relies on a parfor loop. To initialize my parpool I use the command

    parpool(20) % where 20 is the number of cores I have requested.

Occasionally I get the error

    Error using parpool (line 113)
Failed to start a parallel pool. (For information in addition to the causing
error, validate the profile 'local' in the Cluster Profile Manager.)
Caused by:
    Error using parallel.internal.pool.InteractiveClient>iThrowWithCause (line
    675)
    Failed to start pool.
        Error using parallel.Job/submit (line 351)
        Error closing file
        /gpfs/home/cholt/.matlab/local_cluster_jobs/R2017b/Job47.in.mat.
        The file may be corrupt.

Sometimes my code works fine. And sometimes I get that error. I don't really understand what's happening and why that error appears sometimes but not always. How can I fix it? Thanks for your help -C

Answer 1

[Zenin Easa Panthakkalakath]

Hey Caleb,

It is possible that one or more of the workers never managed to fully start. This may have been caused by certain preference settings. Try to regenerate the MATLAB preferences.

Also can you please check the "startup.m" file if it contains any commands which will alter the MATLAB preferences. If yes, try commenting out these and run the program again.

Regards

Zenin

[Caleb_Holt]

Hey Zenin

I haven't messed with the preference settings at all, that file is still empty. I also don't have a startup.m file. I think the problem is that it is trying to access a job file that doesn't exist. I'm not sure if I should make those files or what needs to happen.

Answer 2

[Zenin Easa Panthakkalakath]

Hey Caleb,

I understand that you haven't made any changes to the preference settings. In that case, it could mean that the MATLAB Cluster may not have been validated. Please refer to these pages in order to do the same:

• Validate cluster profile - MATLAB Documentation page
• Manually validate cluster profile - MATLAB Answers page

Please let me know if this works for you.

Regards

Zenin

[Caleb_Holt]

So Zenin, the problem I believe has nothing to do with creating the parpool itself. Rather, it has to do with where the job files are being created which is under my name directory which has limited space. I would like to change the location those files are created and saved. How do I do that?

[Caleb_Holt]

In particular these lines of the error message

Error using parallel.Job/submit (line 351) Error closing file /gpfs/home/cholt/.matlab/local_cluster_jobs/R2017b/Job47.in.mat. The file may be corrupt.

I think the fact that I've reached my space limitation on my home directory, where the .matlab file is stored, is what's throwing this error. How can I change where that .out.mat file is stored?

[Zenin Easa Panthakkalakath]

Hey Caleb,

The directory that you've mentioned is the preference directory.

>>prefdir % this would return the preference directory.

In order to set a custom preference directory, please try the solution in this article .

Regards

Zenin