You can control for cases where the oozie launcher uses all your task slots on your cluster by creating a dedicated oozie_launcher queue for on your jobtracker and then adding the following configuration property to the offending action is your oozie workflow.
<property> <name>oozie.launcher.mapred.job.queue.name</name> <value>oozie_launcher</value> </property> On Sun, Aug 18, 2013 at 3:51 PM, Harsh J <[email protected]> wrote: > Are you certain your YARN cluster has adequate resources/proper > configuration to run both of these jobs simultaneously. A hang in > ACCEPTED state for a submitted MR job typically means its started, but > lacks adequate resources to schedule further tasks/containers. > > The launcher job, i.e. the "oozie job name" job, is the parent java > class invoking job which will wait around until the java application > truly finishes. In case of the java app being an MR job itself, this > means "until the job is complete". > > On Thu, Aug 8, 2013 at 10:16 AM, Voronov, Dmitry <[email protected]> > wrote: > > I have workflow job with a java action node. Run with Hadoop > 2.1.0.2.0.4.0-38 and Oozie 3.3.2.2.0.4.0 When I submit the job I see 2 > lines in Hadoop Resource Manager screen. 1. with original job name 2. with > Oozie job name. > > > > The task with Oozie job name is hanging in "RUNNING" state The task with > original name is in "Accepted" state. All that I see in logs is: > > > > >>> Invoking Main class now >>> > > Heart beat > > Heart beat > > Heart beat > > Heart beat > > ... > > > > > > Thank you > > > > -- > Harsh J >
