Split Vote
An election in which no candidate obtains a majority, so the term ends with no leader and the process must repeat.
In majority-based election, several nodes can time out at nearly the same moment, all become candidates, and split the votes so that none reaches a majority. No leader is elected, the term ends, and the cluster is unavailable for writes until a new election succeeds.
Without a mitigation this can repeat: the same nodes time out together again and split again. The standard fix is randomised election timeouts — each node waits a random duration within a range, so one usually times out clearly first, becomes a candidate, and collects votes before the others wake. Raft relies on this, and it is why the range must be comfortably larger than the typical round-trip time.
Two operational consequences. Election timeouts that are too short relative to network latency cause frequent spurious elections and repeated split votes — a cluster that keeps changing leader under load is usually misconfigured this way. And even-sized clusters are worse, because a tie is possible; this is the reason cluster sizes are odd and why a witness node exists.