Calls to RPC.stopClient render RPC subsequently unusable – at least until a reload of RPC class so static initializer has a chance to run again.
I am trying to write unit tests for a little cluster built atop hadoop RPC class. Intent is to spin up the little cluster before each test is run. Post unit test, the cluster is torn down. Part of the takedown includes invocation of RPC.stopClient to clean up any outstanding connections, etc.
I've found that when the second unit test runs, RPC Client is horked. The 'running' flag is false so Connections can't work and 'Connection Culler' is not running.