- proper logging with swift-log
- use the
Nodeinterface of swift-cluster-membership - implement tests for the Paxos implementation
- test more edge cases in
ClusterTests - implement graceful leave
- implement timeout-based failure mechanism (right now we rely solely on the multi-node cut detection which is more of a fast-path)
- proper
SortableSetimplementation, now it's mosly a hack that will perform poorly for large clusters