Leader election allows a group of processes to select one coordinator. Classic algorithms include Bully (highest ID wins via comparison messages), Ring (messages circulate), and randomized (Raft uses randomized timeouts). All algorithms must ensure at most one leader is elected, handle leader failures, and elect a new leader when needed.
Many distributed systems simplify coordination by designating a single node as the leader (or coordinator, or primary). The leader makes decisions — accepting writes, assigning tasks, ordering operations — so that the other nodes do not need to negotiate among themselves for every action. But this creates a problem: what happens when the leader crashes? The remaining nodes need a way to agree on a new leader, quickly and without conflict. This is the leader election problem.
The Bully algorithm is the most intuitive approach. Every node has a unique numeric ID. When a node suspects the leader has failed, it sends an "election" message to all nodes with higher IDs. If any higher-ID node responds, the initiator backs off and lets the higher node take over the election. If no one responds, the initiator declares itself leader and announces this to all lower-ID nodes. The "bully" name comes from the dynamic: the highest-ID node always wins by bullying everyone else into submission. The algorithm is simple but generates a lot of messages — O(n²) in the worst case — and assumes reliable failure detection, which as you know is itself imperfect.
The Ring algorithm takes a different approach. Nodes are arranged in a logical ring, and election messages travel around the ring collecting node IDs. When a node detects the leader has failed, it sends an election message containing its own ID to its successor. Each node along the ring appends its own ID and forwards the message. When the message returns to the initiator (having traveled the full ring), the node with the highest ID in the message is declared leader, and a coordinator announcement circulates to inform everyone. This uses fewer messages than the Bully algorithm but is slower because it must complete a full ring traversal.
Modern systems often use randomized approaches. The Raft consensus algorithm, which you will study next, uses randomized election timeouts: when a node detects the leader is missing, it waits a random amount of time before starting an election. This randomization makes it unlikely that two nodes start elections simultaneously, avoiding most split-vote scenarios without complex message exchanges. The critical safety property across all these algorithms is that at most one leader exists at any time — a situation with two leaders (a "split brain") leads to conflicting decisions and data corruption. Achieving this guarantee in the presence of network partitions and asynchronous communication is what makes leader election genuinely difficult, and why it connects deeply to the broader consensus problem.