Update cpp_fmriprep.slurm

[marcobarilari]

Do these numbers make sense?

[marcobarilari]

They seem to the resources limit for lemaitre3 (deducted from https://www.ceci-hpc.be/scriptgen.html) and # SBATCH --ntasks-per-node=2 from Filippo tests does not seems to make any difference

[yyang1234]

Hi, for me my problem isn't related to memory but I tried to run with these numbers and they works. So they make sense.

[marcobarilari]

If with these numbers you were able to run to the end without error the problematic subjects THEN (1) it is very good news and (2) very likely that your problem could not find file X was indeed related to memory.

Possible explanation In a nutshell: we provided too little "ram" memory so that fmriprep was forced to free it up during some processes, therefore it deleted cached files that were supposed to be used later and not being there it dumbly through the error you saw.

LMK if it makes sense to you

[yyang1234]

Hi, what you said is an important point. But the error that happened to me appeared even with these new numbers. The error was solved by clearing the content in both the output folder and the scratch folder.
Now in the slurm file I define workdir like this --work-dir /scratch_dir/work-fmriprep/"${subjID}" \ instead of --work-dir /scratch_dir/work-fmriprep \ to make the scratch folder less likely to be mixed with each others. Does it make sense?

[yyang1234]

I also found that for instance, a single subject has three tasks, running the first task works without any issues. However, when I subsequently run the second task for the same subject, an error occurs. Now I run all tasks for one subject at one time to avoid the issue.

[marcobarilari]

Ah interesting, well this could be important to change as well in this repo. Will run some trials.

[marcobarilari]

I also found that for instance, a single subject has three tasks, running the first task works without any issues. However, when I subsequently run the second task for the same subject, an error occurs. Now I run all tasks for one subject at one time to avoid the issue.

What was the error? Let's share it in the trouble shooting section in his repo

[yyang1234]

The same error as I posted before. I think the error related to recycled either the output folder or the scratch folder within same participants.