Hey ZooKeepers -
A while back we patched in the awesome ``mntr`` change, and started
publishing those stats to Ganglia. They're super handy :D
Looking at those graphs we see connections are not evenly distributed
among cluster members, where some have thousands of connections and
others are largely idle. Any ideas about how to more evenly distribute
connections? In practice this leads to overloaded instances, session
timeouts, pages to operations, etc.
Already filed a jira so I could attach some graphs:
https://issues.apache.org/jira/browse/ZOOKEEPER-856
Has anyone else seen this?
--travis
