symat commented on issue #1254: ZOOKEEPER-2164: Quorum members can not rejoin after restart URL: https://github.com/apache/zookeeper/pull/1254#issuecomment-588113917 During startup, everyone starts a channel to everyone else roughly the same time (e.g. when I did with docker-compose). So there will be N * (N-1) connections initiated. Half of them will be closed (as the challenges gets lost), but the other half will kept open and there will be no need for 'sending back' initial messages and no wrong address will be used. Also if you start the nodes sequentially in the order of 'myid' (most probably this is quite usual in production for the initial startup), then the problem will not be seen. In this case the 'new server' will start to initiate connections to the old ones, and he will always win the challenge (his server id is always larger).
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
