What happens to the server and its clients if one of the servers in the zookeeper cluster has lost its neighbors - it continues to give out the data that was at the time of connection loss (like etcd), disconnects all its clients, continues to give data except for ephemeral nodes or something something else?

Context of the question: you need to have access to relevant data. If this is not possible, find out in time that there is no access and turn off. In the etcd cluster, if the server has lost the connection to the common cluster - it continues to give data, but writing to it is impossible. However, it ignores the parameter of the lifetime of the value (ttl), i.e. To determine that the data is relevant, each client must constantly write to the server and if the recording fails - reconnect to another (manually, there is no built-in mechanism). How does it work in zookeeper?

    1 answer 1

    Based on this discussion , this page, and the generally accepted use of the quorum model — that part of the cluster that cannot gather a quorum goes into the waiting state in which it stops accepting customers and only waits for the quorum to recover:

    As a result of the service anymore, it can be used to read. Clients will simply receive a ECONNREFUSED failure at this point. There's no 'read-only' mode - every operation requires a presence of an quorum. The ZK overview doc mentions this implicitly http://archive.cloudera.com/cdh5/cdh/5/zookeeper/zookeeperOver.html :

    "" "
    ZooKeeper service will be available.
    "" "