Job Submit

One of our primary goals is to make running Twister2 jobs as simple as possible. If you downloaded a twister2 distribution, you just need to unpack that tar file.

If you compiled the source code, following tar package is generated under twister2 directory:

bazel-bin/scripts/package/twister2-0.4.0.tar.gz

Unpack the downloaded or generated tar file:

$ tar xf bazel-bin/scripts/package/twister2-0.4.0.tar.gz

Submitting a Job

Twister2 jobs are submitted using the twister2 command. This command is found inside the bin directory of the distribution.

Here is a description of the command

twister2 submit cluster-type job-type job-file-name job-class-name [job-args]
  • submit: the command to execute
  • cluster-type: either of standalone, kubernetes, nomad, slurm, mesos
  • job-type: at the moment, we only support jar
  • job-file-name: the file path of the job file (the jar file)
  • job-class-name: name of the job class with a main method to execute

Here is an example command:

./bin/twister2 submit standalone jar examples/libexamples-java.jar edu.iu.dsc.tws.examples.task.ExampleTaskMain -itr 80 -workers 4 -size 1000 -op "allgather" -stages 8,1

Terminating a job

To kill a job, ```twister2 kill`` command is used.

Here is a description of the command:

twister2 kill cluster-type <job-id>
  • kill: the command to execute
  • cluster-type: either of standalone, kubernetes, nomad, slurm, mesos
  • job-id: id of the job to kill

Note: when jobs complete, they are terminated automatically. Kill command is used to kill jobs that does not terminate.

Listing jobs

When ZooKeeper is used to store job metadata, this data stays at ZooKeeper server even after the job has completed. Users can query ZooKeeper server and see the status of current and past jobs.

Here is a description of the command:

twister2 list <jobs/job-id>
  • list: the command to execute
  • jobs/job-id: if the word "jobs" is given as the third parameter, current and past jobs are listed. If the job-id is given as the third parameter, then the status of that job is listed with its workers.

For this command to work, ZooKeeper address has to be provided as the value of following parameter in the common/resource.yaml file:

twister2.resource.zookeeper.server.addresses

Listing Jobs in Kubernetes Clusters

When running jobs in Kubernetes clusters, sometimes ZooKeeper server may run in Kubernetes cluster and there may not be a direct connection from user machine to ZooKeeper server. In such cases, users can run the list command in the cluster by creating a twister2 pod and executing the commands in that pod.

We have designed a sample twister2 pod for this purpose. You can create that pod as:

kubectl create -f https://raw.githubusercontent.com/DSC-SPIDAL/twister2/master/twister2/config/src/yaml/conf/kubernetes/deployment/twister2-pod.yaml

After creating the pod, you can set your ZooKeeper address at the config file as:

kubectl exec -it twister2-pod -- nano conf/common/resource.yaml

You can list the jobs as:

kubectl exec -it twister2-pod -- bin/twister2 list jobs

You can list a particular job assuming the job-id as: username-t2-job-gnrk4yt

kubectl exec -it twister2-pod -- bin/twister2 list username-t2-job-gnrk4yt

Do not forget to delete the pod after you are done with it:

kubectl delete -f https://raw.githubusercontent.com/DSC-SPIDAL/twister2/master/twister2/config/src/yaml/conf/kubernetes/deployment/twister2-pod.yaml
← Python APIStandalone →