I’ve started seeing this after I moved to 1.1, and it’s causing me aggravation. Perhaps someone can shed some light on it.
If a topology is active in ZK, but no nimbuses have the code, then none of the nimbuses can be the leader, and then the whole cluster won’t start. So, I have to manually remember to kill all running topologies before taking down my cluster, and If I don’t, it seems the only alternative is to clobber /storm in zookeeper. I inevitably forget, and then it’s a real pain. Surely, there must be a better way? Can we assume that if a topology is active in ZK, but no nimbuses have the code, then that means the whole cluster died and so we should deactivate the topology in ZK? Is there some good reason for this change that I don’t see? Thanks Dan
