The clustering technologies are based on the raft protocol. The protocol can ensure failure tolerance if there are enough backup nodes to cover for the failure. To cover for the failure, a consensus is made using leader election. The election can only work if there is more than half the nodes electing the same leader.

With only two nodes, if one node fails, the other node is left  with only one vote (its own vote) and can't continue to be the leader.

To put it another way, two nodes are enough if one node fails entirely, but cannot cover the case where both nodes are running but the network is failing between them. Each node will think the other failed and no node will know who is really in charge.

See also how raft works in consul.

Did this answer your question?