Sorry, Rohith Sharma K S, I started working on this one before I saw you'd posted a patch.
The issue is timing, as is the case with almost all our flakey tests. The node report is requested before the cluster has had time to fully initialize. At least that's my theory. Attached is a patch that retries the node report until at least one node is running. (Since there's only one node requested, that's all that's needed.)
I'm trying to track down another test failure in AMRMClient, so I'll run my tests with this patch applied and see if the ArrayIndexOutOfBoundsException shows up again.