Open jbwexler opened 3 weeks ago
could you remind how container executed in your case? if via our singularity_cmd
then we might need to make sure that we have ~/.cache
in our "fake HOME". but also we might need to fix datalad -- it should create all leading folders there and not assume ~/.cache
to be present.
I'm using ReproMan to submit a slurm job on TACC's Frontera. I also add --bind /tmp:/node_tmp
into the singularity command so that it will use the node's local temp rather than the scratch temp. I add that into the containers/.datalad/config file so it runs that way automatically when I submit via reproman. Here's an example of a command-array file:
code/containers/scripts/singularity_cmd run --bind /tmp:/node_tmp code/containers/images/bids/bids-mriqc--24.0.0.sing sourcedata/raw /scratch1/03201/jbwexler/openneuro_derivatives/derivatives/mriqc/ds004488-mriqc participant --participant-label '01' -w '/node_tmp/work_dir/mriqc/ds004488_sub-01' -vv --nprocs 11 --ants-nthreads 8 --verbose-reports --dsname ds004488 --mem_gb 30
@yarikoptic Any ideas for a workaround?
Getting this for all MRIQC runs for any dataset: