Details
-
Task
-
Resolution: Unresolved
-
Critical
-
Cheshire-Cat
-
1
Description
per comment here: http://review.couchbase.org/c/ns_server/+/144677
I still think though that we shouldn't cast in the first place. It's got to be easier to synchronize with mb_master locally. For one thing, the process calling shun() could trap exits to avoid being interrupted by mb_master shutting down leader processes. In itself that wouldn't be enough because the moment other nodes see the change to nodes_wanted, they'll stop accepting heartbeats from the old leader. But that could probably be dealt with by ordering the updates carefully. In any event it seems simpler than asking a second node to wait for a third node to become the new leader and complete the job.
I guess we can come back to this when non-atomicity of chronicle topology/nodes_wanted updates is dealt with.