The distributed system can only handle one point of failure, and two have failed.

The leader is net split and doesn't care about most of the cluster and the "zookeeper" is happy with the leader.

If the zookeeper doesn't select a new leader the cluster is going to stay in this state.