With the change to isAnyNodeOk() in DBClientRS::connect(), we rely on the ReplicaSetMonitorWatcher to keep us up-to-date (if it's stale, our connections will fail). This behavior is different from previous, where we actively checked the replica set again for activity.

Think if we aren't able to find nodes we know are ok, we should re-check the nodes. A timed backoff may be appropriate as well to avoid spamming reconnect requests.