ds000007-catqc/code/participant_job

75 lines
2.8 KiB
Bash
Executable file

#!/bin/bash
# the job assumes that it is a good idea to run everything in PWD
# the job manager should make sure that is true
# fail whenever something is fishy, use -x to get verbose logfiles
set -e -u -x
dssource="$1"
pushgitremote="$2"
# if dataset has sites subd includes it
subd="$3"
subid=$(basename $3)
# get the analysis dataset, which includes the inputs as well
# importantly, we do not clone from the lcoation that we want to push the
# results too, in order to avoid too many jobs blocking access to
# the same location and creating a throughput bottleneck
datalad clone "${dssource}" ds
# all following actions are performed in the context of the superdataset
cd ds
# in order to avoid accumulation temporary git-annex availability information
# and to avoid a syncronization bottleneck by having to consolidate the
# git-annex branch across jobs, we will only push the main tracking branch
# back to the output store (plus the actual file content). Final availability
# information can be establish via an eventual "git-annex fsck -f catqc_out-storage".
# this remote is never fetched, it accumulates a larger number of branches
# and we want to avoid progressive slowdown. Instead we only ever push
# a unique branch per each job (subject AND process specific name)
git remote add outputstore "$pushgitremote"
# all results of this job will be put into a dedicated branch
git checkout -b "job-${JOBID}"
# we pull down the input subject manually in order to discover relevant
# files. We do this outside the recorded call, because on a potential
# re-run we want to be able to do fine-grained recomputing of individual
# outputs. The recorded calls will have specific paths that will enable
# recomputation outside the scope of the original Condor setup
datalad get -n "sourcedata/raw/"
# the meat of the matter
# look for T1w files in the input data for the given participant
# it is critical for reproducibility that the command given to
# "containers-run" does not rely on any property of the immediate
# computational environment (env vars, services, etc)
find \
sourcedata/raw/${subd} \
-name "${subid}*T1w.nii.gz" \
-exec sh -c '
odir=$(echo {} | cut -d / -f3);
datalad containers-run \
-m "Compute CAT12.9 QC for $odir" \
-n cat12-9\
--explicit \
-o $odir \
-i {} \
sh -e -u -x -c "
mkdir -p {outputs[0]} || true \
&& cp {inputs[0]} {outputs[0]} \
&& /singularity -b /batch/CATpreprocessing4qc+ROIs.m {outputs[0]}/*.nii.gz \
&& rm -f {outputs[0]}/*.nii* {outputs[0]}/report/*.jpg \
&& gzip {outputs[0]}/*/*.nii \
" \
' \;
# file content first -- does not need a lock, no interaction with Git
datalad push --to catqc_out-storage
# and the output branch
flock --verbose $DSLOCKFILE git push outputstore
echo SUCCESS
# job handler should clean up workspace