I have a Jenkins pipeline job that does the following:

• Create a list of jobs, each of which is a single shell operation which takes a non trivial amount of time
• Run the jobs in the list in parallel

If the jobs take, for example, 10, 5, 7 and 8 minutes then I would expect this stage of the job to take about 10 minutes to complete (the duration of the longest single job with a little extra time for setup and teardown of the parallel operation). However I see something much closer to 30 minutes. A cut-down version of the Jenkinsfile which still shows the problem is below.

node('docker-centos7') {
    stage('Build') {
        sh "date"
        jobs = [:]
        ["foo", "bar", "baz", "qux"].each { vals ->
            jobs[vals] = { sh "date +start_${vals}_%F_%T; for i in \$(seq 1 600); do sleep 1; date +${vals}_%F_%T; done; date +done_${vals}_%F_%T;" }
        }
        parallel jobs 
        sh "date"
    }
}

This should start 4 parallel jobs which each take 10 minutes but on my server this takes > 30 minutes to run. The output of this job on my server clearly shows that there is a lengthy delay between the last job completing and the final sh "date" call.

+ date +done_baz_%F_%T
done_baz_2024-02-28_15:23:57
[Pipeline] }
[Pipeline] }
[Pipeline] }
[Pipeline] }
[Pipeline] // parallel
+ date
Wed Feb 28 15:44:48 GMT 2024

I have some evidence that this might be related to the amount of output that the job produces, if I replace the loop that prints 600 messages with a 1 second delay between each with a single 600 second wait then the extra delay goes away.