We have three topologies running in a single Storm 0.8.1 cluster.
Every few days tuples will stop emitting from bolts and transferred to
bolts in one of the topologies. Storm UI shows Acked and Failed counts
as 0 in Topology Stats. It is as if the communication between bolts
freezes, but there are no error messages in zookeper, supervisor or
worker logs. We suspect the problem might be with ZeroMQ, but don't
know how to confirm this since we could not find a log for ZeroMQ. The
only way to "unfreeze" the topology is to resubmit it, restarting
supervisor did not help. Thanks for your help!
Grokbase › Groups › Google › storm-user › December 2012