SGE and matlab_metadata.mat

I have the following problem with running distributed jobs under Sun Grid Engine (MATLAB R2010b). When I submit multiple jobs and one of them finishes, it gets deleted from the cluster, which is fine. What is not fine, is that the matlab_metadata.mat file, which is common to all the jobs is deleted too, which makes the remaining jobs to fail. How can I fix it?

many thanks

Answer 1

[Ayush]

Hi Marcin,

I understand you are encountering a known issue with distributed jobs in MATLAB R2010b under Sun Grid Engine (SGE). The shared "matlab_metadata.mat" file is being deleted when one job finishes, causing the remaining jobs to fail.

Actually as per my understanding, in MATLAB R2010b, each worker cleans up job-related files upon completion, and the logic does not account for other workers potentially still needing the shared "matlab_metadata.mat" file. This is a race condition and file lifecycle management issue in older MATLAB releases.

You can try the following solutions:

1. You can try using a separate "JobStorageLocation" for each job. Try to ensure that each job has a dedicated folder for its metadata. This way, jobs won't interfere with each other's files.

2. You can try modifying the cleanup behavior if its possible in your workflow.

3. You can try upgrading the MATLAB versions as this issue is fixed in later releases of MATLAB (starting from R2012a and onward).

Hope it helps!