Allow client to connect to cluster with missing hosts
tobgu opened this issue · 1 comments
This is basically a port of this commit:
z-division@4f30e43
The same issue was earlier found in the java client, ZOOKEEPER-1576.
Or is there a reason why this fix has not been ported from the z-division fork to this fork?
I like to bring attention to this issue and opt to port those changes. Our use case is that we have Kubernetes deployment of zookeeper and our application that uses this zookeeper client. During upgrade descheduled zookeeper pods are deleted from kubernetes DNS service and added back after pods start up. During this period it is likely that our service will be restarted, especially that we are applying our Kubernetes resources at the same time for zookeeper and our application, which causes that application is unable to connect to zookeeper during its whole upgrade cycle (one-by-one pod upgrade). This change would drastically decrease our upgrade time.