Fmriprep partial success when accounting for SDC

1

Hey,
So I am at a point where I need help with fmriprep. Please note, I am running it with Singularity and have run it successfully many times. However I have discovered I needed to add the “IntendedFor:” field into my fmap json files in the BIDS data(which has passed bids validator), to account for SDC. After doing this I tested it and it HAS run succesfully. After running a batch script to get all subjects done, most of the subjects are partially finishing (the ~/func directory and the report sub-X.html is not being completed). When looking at my error output the strange thing is the errors aren’t consistent, sometimes I am getting out of memory errors(which I highly doubt because the memory allocated for this is beyond sufficient and because some subjects run successfully) but for instance that error looks like…

OSError: [Errno 28] No space left on device: '/home_dir/projects/niblab/bids_projects/Experiments/Bevel/fmriprep/sub-009/fmriprep_wf/single_subject_009_wf/func_preproc_task_prob_run_4_wf/bold_confounds_wf/add_header/_node.pklz'

or I get something like…

FileNotFoundError: No such file: '/home_dir/projects/niblab/bids_projects/Experiments/Bevel/fmriprep/sub-005/fmriprep_wf/single_subject_005_wf/func_preproc_task_prob_run_1_wf/bold_mni_trans_wf/bold_to_mni_transform/vol0112_xform-00112.nii.gz'

and for reference the command looks like…

singularity exec -B /:/home_dir /projects/niblab/bids_projects/Singularity_Containers/fmriprep.simg \
fmriprep /home_dir/projects/niblab/bids_projects/Experiments/Bevel/BIDS /home_dir/projects/niblab/bids_projects/Experiments/Bevel/fmriprep/sub-001 \
    participant  \
    --participant-label sub-001 \
    --fs-license-file /home_dir/projects/niblab/bids_projects/freesurfer/license.txt \
    --fs-no-reconall \
    --omp-nthreads 16 --n_cpus 16 \
    --ignore slicetiming  \
    --bold2t1w-dof 12 \
    --output-space template --template MNI152NLin2009cAsym \
    --debug \
    -w /home_dir/projects/niblab/bids_projects/Experiments/Bevel/fmriprep/sub-001 \
    --resource-monitor --write-graph --stop-on-first-crash

Like I mentioned above, this command had worked before I modified the json files and even after the modification some of my subjects have run successfully(and I cannot notice a difference in the BIDS data between successful and unsuccessful runs). That being said I do not think it is a singularity or BIDS issue. My next guess is possibly a package issue or possibly something I may need the cluster administrators to adjust but I am not sure. I am really at a loss here, any help is appreciated. Thank you.

~Nichollette

2

Hi Nikki,

The error OSError: [Errno 28] No space left on device suggest the there is not enough hard disk space in the working directory specified via -w /home_dir/projects/niblab/bids_projects/Experiments/Bevel/fmriprep/sub-001. The second error could be related to the inability to write files. You can check disk space on unix systems via df -h, but your mileage might vary due to different ways disk quotas are implemented on multi-user clusters. This also could’ve been a temporary disk space shortage or filesystem malfunction on your cluster.

BTW on HPCs we recommend using local scratch (for example /tmp) for the working directory - it’s usually faster than the mounted network file systems. On some systems it is unfortunately not big enough.

Hope this helps!

Best,
Chris

3

Ahh, okay that makes sense. Thanks for the fast reply Chris I’ll mess around with this!!

1 Like