We have some ZK clusters that service 10000 clients, using ZK for group
membership, leader election, sharding, etc... Additionally we have
clusters that service multi-tennant instances (more than one service
using a ZK cluster). Add on top of that the poor performance of the GC
in 1.6 jvms and more memory is critical/necessity.
Lei Zhang wrote:
Thanks for your reply Mahadev -
I wonder if you can elaborate on how you get to your observation of most
applications that use ZooKeeper needing JVM heap size > 64M. Our platform
uses ZooKeeper to manager cluster of hundreds of nodes with -Xmx32m. The
only time I observed OutOfMemory issue is when my application goes into
endless loop injecting messaging into a ZooKeeper queue.
Allowing ZooKeeper server to take up 3G out of 4G physical memory seems