I am preprocessing 3T data from HCP project on our HPC. It has 1 T1, 1 T2, 4 resting state scans and 2 fieldmaps. My preprocessing output includes 91k cifti. The working directory gets very large for each subject at around 100GB and even bigger. Then fmriprep crushes with message system does not have space. Is it normal for the working dir in fmriprep to get this big? Is there a way to set up a clean up process or a certain flag to control the size of work dir? I am trying to preprocess data in parallel by spinning up multiple containers.
FMRIprep does not have any flags to control the size of the working directory. Make sure you are not using the --low-mem flag, as that will increase the amount of workdir outputs.
Also, why not just use the already preprocessed HCP data?
The only way to control working directory size currently is to restrict yourself to subsets of processing at a time.
One of our current efforts is to move to a mode where we generate a minimum set of derivatives (transforms and reference volumes, mostly), such that all other derivatives can be generated deterministically after the fact. This will dramatically reduce the workflow size as well, leading to a smaller working directory.