We use ZooKeeper for HBase and HBase replication has leaved too many HLogs which are recorded in ZooKeeper. Now we have more than 1M watches and the the average latency from `mntr` is almost 3000. The ZooKeeper cluster seems to be unstable and the client occasionally fail to set data on znodes.
How can I reduce the latency from ZooKeeper? Is it mainly related to the number of znodes?
