When switching over an application-node to a new statsd server, we observed flapping graphs in graphite. How to fix this ...
A few days ago we decided to introduce a second statsd server into our network. After switching over a few application-nodes to the new statsd, we noticed strange graphs in graphite: The graphs for the switched over nodes were flapping between zero and the normal expected value.
We later found out the reason: The first statsd still kept the keys in-memory even if there weren't any new events and thus reported the keys with 0 to graphite.
=> Restart statsd servers when switching nodes to another statsd server.