rerun failed legofit jobs from job array at MetaCentrum (PBS Pro)

legofit_jobarray.sh

#!/bin/bash
#PBS -l select=1:ncpus=64:mem=1gb,walltime=24:00:00

## GO TO WORKDIR
cd $PBS_O_WORKDIR || exit

## EXPORT PATH
export PATH="/storage/brno2/home/jena/bin:$PATH"
export PATH="/storage/brno2/home/jena/miniconda3/bin:$PATH"

## LOAD GSL module
#ml GSL/2.7-intel-compilers-2021.4.0 # IT4I
module add gsl/gsl-2.5-intel-19.0.4-lnarr3s #gsl/2.5-intel-17.0.1 # MetaCentrum

## MODEL
## qsub -v datdir=hcg2_atgc,model=fg03v,stage=1 <job script>
mkdir $datdir/$model

## DATA / BOOTSTRAP
# get individual tasks from tasklist with index from PBS JOB ARRAY
if [ "$rerun" ] # https://stackoverflow.com/a/3601734
then
	TASK=$(sed -n "${PBS_ARRAY_INDEX}p" $PBS_O_WORKDIR/$datdir/$rerun)
else
	TASK=$(sed -n "${PBS_ARRAY_INDEX}p" $PBS_O_WORKDIR/$datdir/tasklist)
fi

boot=$(basename -s .opf $TASK)

outpref=${datdir}/${model}/${model}_${stage}_${boot}

# copy input file and executable to scratch
# cp $PBS_O_WORKDIR/$TASK input ; cp $PBS_O_WORKDIR/myprog.x .

## LEGOFIT
# execute the calculation
# ./myprog.x < input > output

## STAGE 1 or 3
if [ $stage == 1 ] || [ $stage == 3 ]
then
	legofit -t 64 -1 -S 5000@10000 -S 1000@100000 -S 1000@1000000 \
		--stateOut ${outpref}.state \
		${model}.lgo $datdir/$TASK \
		> ${outpref}.legofit
		
		# --stateOut ${model}/${model}_${stage}_${boot}.state \
		# > ${model}/${model}_${stage}_${boot}.legofit

## STAGE 2 or 4
## FIXME
elif [ $stage == 2 ] || [ $stage == 4 ]
then
	prevstage=$(expr $stage - 1)
	stateInName=${model}/${model}_${prevstage}
	# TEST "--stateIn ${stateInName}_boot"{0..49}".state"
	legofit -t 80 -1 --tol 2e-5 -S 1000@2000000 \
		${model}.lgo data/${boot}.opf \
		--stateIn ${stateInName}_data.state \
		$(echo "--stateIn ${stateInName}_boot"{0..49}".state") \
		> ${model}/${model}_${stage}_${boot}.legofit

else
	echo "Wrong stage" && exit 255
fi

# copy output file to submit directory
#cp output $PBS_O_WORKDIR/$TASK.out
#  --stateIn ${stateInName}_boot1.state --stateIn ${stateInName}_boot2.state --stateIn ${stateInName}_boot3.state --stateIn ${stateInName}_boot4.state --stateIn ${stateInName}_boot5.state --stateIn ${stateInName}_boot6.state --stateIn ${stateInName}_boot7.state --stateIn ${stateInName}_boot8.state --stateIn ${stateInName}_boot9.state --stateIn ${stateInName}_boot10.state --stateIn ${stateInName}_boot11.state --stateIn ${stateInName}_boot12.state --stateIn ${stateInName}_boot13.state --stateIn ${stateInName}_boot14.state --stateIn ${stateInName}_boot15.state --stateIn ${stateInName}_boot16.state --stateIn ${stateInName}_boot17.state --stateIn ${stateInName}_boot18.state --stateIn ${stateInName}_boot19.state --stateIn ${stateInName}_boot20.state --stateIn ${stateInName}_boot21.state --stateIn ${stateInName}_boot22.state --stateIn ${stateInName}_boot23.state --stateIn ${stateInName}_boot24.state --stateIn ${stateInName}_boot25.state --stateIn ${stateInName}_boot26.state --stateIn ${stateInName}_boot27.state --stateIn ${stateInName}_boot28.state --stateIn ${stateInName}_boot29.state --stateIn ${stateInName}_boot30.state --stateIn ${stateInName}_boot31.state --stateIn ${stateInName}_boot32.state --stateIn ${stateInName}_boot33.state --stateIn ${stateInName}_boot34.state --stateIn ${stateInName}_boot35.state --stateIn ${stateInName}_boot36.state --stateIn ${stateInName}_boot37.state --stateIn ${stateInName}_boot38.state --stateIn ${stateInName}_boot39.state --stateIn ${stateInName}_boot40.state --stateIn ${stateInName}_boot41.state --stateIn ${stateInName}_boot42.state --stateIn ${stateInName}_boot43.state --stateIn ${stateInName}_boot44.state --stateIn ${stateInName}_boot45.state --stateIn ${stateInName}_boot46.state --stateIn ${stateInName}_boot47.state --stateIn ${stateInName}_boot48.state --stateIn ${stateInName}_boot49.state \

rerun_failed_legofit.sh

#!/bin/bash

## bash rerun_filed_legofit.sh hcg2_tv/1Bc

path=$1 # hcg2_tv/1Bc/ (data/model)
files=${path}/*.legofit

## GET PARAMS
datdir=$(echo $path | tr "/" "\t" | cut -f 1)
model=$(echo $path | tr "/" "\t" | cut -f 2)
#echo $datdir $model

stage=$(grep -c DiffEv $files \
| gawk -F":" '
$2 == 0 {
	split($1, arr, "/+") # make sure there really are 3 parts

	if (length(arr) != 3) {
		print "check your path!"
		exit 1
	}

	data = arr[1]
	model = arr[2]
	file = arr[3]

	split(file, arr2, "_")
	stage = arr2[2]
	opf = gensub(/legofit/, "opf", 1, arr2[3])

	## format: data/boot1.opf > hcg2_tv/1Bc_1.tasklist
	print "data/" opf > data "/" model "_" stage ".tasklist"
}
END {
	print stage
}
')

rerun=${model}_${stage}.tasklist
#echo $rerun

if [ -f ${datdir}/${rerun} ]
then

	tasks=$(grep -c '' ${datdir}/${rerun})
	
	## SUBMIT LEGOFIT as job array
	## qsub -A Project_ID -q queue -l select=x:ncpus=y,walltime=[[hh:]mm:]ss[.ms] jobscript
	qsub -N legofit \
		-J 1-$tasks \
		-o $HOME/qlogs \
		-e $HOME/qlogs \
		-v datdir=$datdir,model=$model,stage=$stage,rerun=$rerun \
		legofit_jobarray.sh
	
	#	-W depend=afterany:12623839.meta-pbs.metacentrum.cz \
fi

Story

These scripts are for reruning failed legofit jobs at MetaCentrum. MetaCentrum is a Czech national grid of HPC clusters for scientific computing, and legofit is a package for demographic inference we use to study human evolutionary history.

Legofit is typically run with bootstraps replicates and it's best to submit it as job arrays, using e.g. qsub -J 1-50 in PBS Pro / OpenPBS, the scheduler at MetaCentrum. But due to technical issues at MetaCentrum, some of my jobs failed to finish. These scripts detect which jobs did not finish and resubmit them as new job arrays.

The scripts

• legofit_jobscript.sh - This was my usual script for running legofit as job arrays. The only new part is the if [ "$rerun" ] block that detects if the current job should run as full new job array from standard tasklist or rerun failed jobs from custom tasklist created by the second script.
• rerun_failed_legofit.sh - an absolute hack of a script that takes one argument - a partial path to input data and outputs - and figures out if there are any failed jobs, prepares a list of input (data and/or bootstrap replicates) and submits a new job array with the correct data, model, and stage of legofit analysis. The main work is done with within a gawk script that is brutally grafted into the shell script. The rest of the script was mostly copied from my normal script for submitting such arrays.

Take home message

Learn snakemake FFS 😆