After a partial failover, recovery and rebalance, client would not reconnect

When doing gets in a loop on a 4 nodes cluster, if we failover 2 nodes for a few seconds, recover them and do the rebalance, the client doesn't pick up traffic again.
In fact, no operations at all are seen as soon as the nodes are failed over.

Client ends up failing to pick up the nodes due to a race condition, and the known list of nodes gets out of sync with the newly pushed configuration. This causes a IllegalStateException ("Node not found for request xxx").