Hi,

I have searched the documentation for an answer to the question: At which 
locations does Oozie spawn JVMs when executing a workflow?
Unfortunately, I've only been able to partially answer this question myself. It 
is quite obvious that for MR actions the regular YARN JVM
spawning applies. For Java actions I was able to derive that they are run using 
a single-map YARN job as well.

My question now is:
Is the actual workflow itself run in a separate JVM? If yes, does that occur on 
the node where the Oozie server itself lives? Or is all Workflow
control logic directly executed within the Oozie server process? Besides the 
workflow itself, how are HDFS actions and control nodes treated in
regard to those questions?

Thanks in advance and kind regards,
Michael

Reply via email to