Hi, We tried to perform upgrade of the zookeeper version from 3.5.6 to 3.6.2 version in our ensembles.
We tried to perform this using both rolling-upgrade and also by bringing down the ensemble and upgrading all the nodes in the ensemble. I see that after around 10 minutes, I see the below messages pop up in logs and the zookeeper server stops responding to the requests. My zookepeer configuration looks like below. During this time I see that close to 2000 connections are in the CLOSE_WAIT state on each zookeeper server and on the client these connections are in the FIN_WAIT_2 state. I suspect zookepeer server is not able to close the connections quicky. Eventually the number of open FD will hit the limit configured(4096) and fail as they cannot synchronize with the leaders. We see that the 3.5.6 version of zookeeper is able to handle the same amount of traffic without causing socket leaks? Could you please suggest what could have gone wrong here. -- Sent from: http://zookeeper-user.578899.n2.nabble.com/
