Occasionally when using the Vanilla 2 Hadoop plugin, a job will fail in the Oozie launcher when it tries to contact the jobhistory server at mapreduce.jobhistory.address. The default value for this parameter is 0.0.0.0:10020, and it needs to be set in mapred-site.xml.
It is unclear exactly why the jobhistory server is only contacted sometimes. The frequency seems to be relatively low, maybe less than 10%. Difficult to tell. I do not have a stack trace handy, but when it fails Oozie reports "Connection refused" to 0.0.0.0:10020 in the job log on the worker node.
Note, the HDP 2 plugin already sets this configuration parameter in a method called finalize_configuration, after services are deployed.
The Vanilla 2 plugin should set this as well. There are two potential ways to do this:
1) Set it in the job configuration, the same way we set values for namenode and datanode on each job. (Of course, this is redundant in the case of the HDP plugin and is irrelevant for Hadoop 1, so we would want to check the hadoop version). Quick fix maybe, but a little kludgy.
2) Make the Vanilla 2 plugin update this config on the cluster once the jobhistory server is launched, as in the HDP plugin. This is better but more complicated.
Fix proposed to branch: master
Review: https:/