increase event handling buffer

this should mitigate a timing issue where joining larger clusters would
cause cluster.members() to deadlock.

cluster.go

@@ -101,7 +101,9 @@ func newCluster(config *config, wg *wgState) (*cluster, error) {
 		localName: ml.LocalNode().Name,
 		ml:        ml,
 		wg:        wg,
-		events:    make(chan memberlist.NodeEvent, 1),
+		// The big channel buffer is a work-around for https://github.com/hashicorp/memberlist/issues/23
+		// More than this many simultaneous events will deadlock cluster.members()
+		events: make(chan memberlist.NodeEvent, 100),
 		state:  state,
 	}
 	mlConfig.Conflict = &cluster