I have a couple of jobs still running in Zeppelin after several days. Cancelling the jobs in Zeppelin has no effect. I've stopped all services, restarted ambari and even rebooted the servers and the jobs won't go away.
When you initiate a interpreter, that starts a yarn application in cluster,
there are couple of ways to clear the sessions
1. Terminating from Yarn (most effective)
list the yarn application find out the application id running from zeppelin and terminate it.
you may use the Yarn UI (running applications --> select the application --> kill the application - right side top corner of app details)
if you have yarn ACLS and kerberized the cluster
obtain the ticket (kinit <user>)
yarn application -list
yarn application -kill <appId>
2. By killing the interpreter processor (from terminal) you can celan-up the PIDs of the Interpreter process in Zeppelin server and kill the pid, wil ensures that AM get aborted(might not be graceful).