Client automatically, failover to other machines when it detect problem. maxClientCnxn is maintained by each server (a simple IP to connection count map)
If you only have m connections(sessions) per IP, then that should be the right value to use. However, it is possible that a given IP can create 3*m sessions to the ensemble in your case. -- Thawan Kooburat On 5/29/13 5:31 AM, "Richards Peter" <[email protected]> wrote: >Hi all, > >Thanks for your replies. > >I will try to describe my setting in very simple terms. Assume that I have >n machines running m number of JVMs each. Each JVM would connect to >zookeeper cluster(having 3 servers currently) to maintain some state >information. In this case would you recommend me to restrict >maxClientCnxns >to m? > >Initially I was thinking that zookeeper cluster is keeping a kind of state >replica in all the machines. But when I read about the description of >maxClientCnxns, I felt that it is not replica which is stored in >zookeeper. >Each client connection could go to a different zookeeper server. So I >wanted to confirm whether the clients would connect in a load balanced >fashion. > >Since clients connect randomly to any server, would it lead to increased >loads on some servers than the others? > >Thanks, >Richards.
