NPE when attempting to recover from NameNode outage

It seems that whenever the NN goes offline, our collectors almost always have problems reconnecting when it's back up. Sometimes sending a 'refreshAll' via the Flume console is enough to fix it; other times we have to bounce the service. The below logs are from one particular instance where we saw a problem.

http://pastie.org/pastes/1918456/text?key=iritbrui0wqf1afq193ew