When running our mr job, sometimes we get an error saying that the Container launch failed for container XYZ: YarnException: Unauthorized request to start container.
A google search says that this is related to ntp and time being out of sync but the time was synced on all the nodes for one test (4 nodes) and another test was only using 1 node (decommissioned from 3 out of 4 nodes).
What else could cause this type of error?
This was with CDH3. Some google results implied that making sure that all the nodes were in sync with respect to time would get rid of the error but that didn't seem to help in this case.
We've since moved to another configuration but I'm still curious if any other ways to avoid that message might work or if it is something that only appears on earlier CDH versions.