Consider five database replicas. All of them can handle reads, but only one should handle writes. The question is: which one?

If multiple nodes accept writes simultaneously, you get conflicting data. If no node is allowed to accept writes, your system stops. You need one valid node to take charge, and everyone else needs a reliable way to recognize that authority.

This is the leader election problem. Network failures, crashed nodes, and stale leaders make it harder than picking a machine.

Getting leader election wrong can cause data corruption, split-brain scenarios, or outages. In interviews, what matters most is explaining how you keep two nodes from acting as leader at the same time, rather than naming every algorithm.