hi,
we meet a issue that the temporary node of broker in zookeeper was lost
when the network bewteen broker and zk cluster is not good enough, while the
process of the broker still exist. as we know, the controller would consider it
to be offline in kafka. After we open zkClient log, we can find when the
connection state between broker and zk cluster is changed from disconnected to
connected, but the newSession callback is not called.so this
broker can not recover to alive except restart.
So we decide to add a heartbeat mechanism in the application layer
between client and broker that distinguish from zkclient heartbeat. Can we
immediately register this broker to zk when we detect broker temporary node is
not in zk path. or how can we solve it?
The main problem is that the watch callback has the possibility of
miss, how can we solve it?
Thanks.
