Hi I'm doing a simple benchmark to record the time for issuing a sbatch command. The contents of the script are: #!/bin/bash
IFS='= ' read _ local_clusterid <<< $(scontrol show config |grep -i clustername) # Extract local cluster name echo "Local cluster: "$local_clusterid # Check input clusters to submit job(s) if [ -z $1 ]; then clusterid=$local_clusterid else clusterid=$1 fi # Check input constraints if [ -z $2 ]; then cons='' else cons=$2 fi # Check input number of tasks if [ -z $3 ]; then no_tasks=1 else no_tasks=$3 fi echo "start time before sbatch "$(date +%s.%N) job_cluster_id=$(sbatch --parsable --wait -M$clusterid --constraint=$cons -n $no_tasks test_srun_with_time.sh) # Since sbatch is a non-blocking operation, --wait is used to block sbatch until response is received. Input arguments are cluster(s) name(s), constraints, number of tasks echo "end time after sbatch "$(date +%s.%N) IFS=';' read jobid clusterid <<< $job_cluster_id # Extract job ID and cluster ID (if not local cluster) echo "JOB ID: "$jobid", clusterid: "$clusterid sacct -j $jobid -M$clusterid --json > "test_"$jobid"_cluster_"$clusterid ".json" # Extract job statistics into a JSON file The "elapsed" time in the resulting JSON file gives me 4 secs, while the time difference between the start and end time in the previous script gives about 10 secs. I wonder why there is much difference between the 2 records? and which one I should use. The output of the previous scripts is: Local cluster: slurmcluster1 start time before sbatch 1688937564.063906087 end time after sbatch 1688937574.160335632 JOB ID: 200, clusterid: slurmcluster1 and the resulting JSON file is attached Regards -- Mohammed
test_200_cluster_slurmcluster1.json
Description: application/json
