Helm and Kubernetes: Is there a barrier equivalent for jobs?

问题

Given 3 jobs (A,B,C) in 3 Helm charts, is it possible to run A and B jobs in parallel, then start job C as soon as both of them are finished? Think of a barrier, in which a bunch of stuff needs to be finished before moving on.

Even if I put A and B charts as sub-charts for C chart, then all 3 are started in parallel.

I already have an workaround for this: add an external check for A and B job finishing, then start C. Still, I would prefer a Helm-based solution, if it exists.

回答1:

Kubernetes isn't a batch job framework/scheduler and does not fit your advanced batch framework requirements.

My recommendation would be to use a real batch framework like Luigi which also supports scheduling Luigi jobs on kubernetes. Look here for an example how to do this.

回答2:

Indeed, Kubernetes seems to be quite basic when it comes to scheduling jobs. We'll move to Luigi at some point in the future for advanced job scenarios.

For now, I wrote this small awk, bash-based workaround for this. Perhaps it could help others in a similar situation.

while true; do
    sleep 60

    # done means output is 'SUCCESSFUL\n1' only
    is_done=$(kubectl get jobs 2>&1 | awk '{print $3}' | uniq | awk 'BEGIN{no_lines=0; no_all_lines=0} {no_all_lines++; if(NR==1 && $1=="SUCCESSFUL") {no_lines++}; if(NR==2 && $1=='1') {no_lines++}} END{ if(no_lines==no_all_lines) {print "1"} else {print "0"}}'
)

    if [ ${is_done} = "1" ]; then
        echo "Finished all jobs"
        break
    else
        echo "One or more jobs are still running, checking again in 1 minute"
    fi
done

来源：https://stackoverflow.com/questions/45454130/helm-and-kubernetes-is-there-a-barrier-equivalent-for-jobs