Hi, Nimbus thinks that the worker where that bolt/spout is stalled or died and kills it, rebalancing the cluster to recover from it.
We are having exactly that behavior on a spout I implemented that is affected by the size of the data it handles and I'm now rewriting it so it doesn't depend on our data size and keep its runtime stable over time to prevent this problem from happening. On 30 June 2015 at 16:31, Casey McQuillan <[email protected]> wrote: > Hello, > > > > We’re currently running storm with a number of multi-lang components. Our > current version is 0.9.3 which, by my understanding, requires the > subprocess to send back a response to heartbeat tuples. My question is, > what is storm’s expected behavior when a subcomponent does not return a > heartbeat? If, say, a single tuple takes an unusual amount of time to > process, causing multiple heartbeats to pile up, how does Storm treat that > subprocess? > -- Carlos Perelló Marínhttps://www.serverdensity.com
