Hello all,

We run a Hadoop cluster of 60 datanodes (0.20.2, CDH3u4). Almost all jobs are 
scheduled via Oozie. Recently I’ve noticed that we get ~200 oozie.log entries 
per minute for actions of killed workflows:

[26/12/2013:16:00:55 PST] pool-2-thread-30  WARN 
org.apache.oozie.command.wf.SignalCommand: USER[oozie] GROUP[users] TOKEN[] 
APP[sv1_prod_client-20131001T1303-kec-W] 
JOB[0063764-131209015949959-oozie-tell-W] 
ACTION[0063764-131209015949959-oozie-tell-W@check-work-directory] Workflow not 
RUNNING, current status [KILLED]

These workflow were in fact killed using ‘oozie job -kill' command. But these 
updates related actions continue forever. Each new oozie kill results in one or 
even two more zombie actions.

Where are these zombies come from? How do we stop them?

Thanks,
mtjr

Reply via email to