I am noticing the following behavior when nodes leave and join a cluster due to network partition. When a set of nodes is disconnected from the cluster, there's is one CPG config change callback with a list of nodes that left the cluster. However when they rejoin the cluster, all of then pretty nearly concurrently as the network segment is reconnected to cluster, there are multiple CPG callbacks each one containing a single node. There's a single Totem config change callback with the list of members that includes all nodes that rejoined the network.
Furthermore when I look at the code, it appears that the list of nodes that joined the cluster is available in message_handler_req_exec_cpg_joinlist() which then calls do_proc_join() in a loop. What's the reason for this asymmetry, i.e, a single CPG notification when nodes leave the cluster, but individual notifications when they rejoin the cluster ?
Sathya
_______________________________________________ discuss mailing list discuss@xxxxxxxxxxxx http://lists.corosync.org/mailman/listinfo/discuss
