split brain
A failure in which two nodes both believe they are the leader and both accept writes.
Two leaders accepting conflicting writes forces repair to discard data users were told was saved.
In more detail
It usually follows a network partition during failover without a fence. Replicas diverge, clients see conflicting answers, and repair must discard some acknowledged data. The fix is a single write authority protected by leases or epochs.
Related terms
- fencing — Cutting a failed or doubtful member of a replicated system off from its data so it cannot keep writing while you recover.
- leader epoch — A counter per partition that goes up by one every time a new leader is elected.
- network partition — A network failure that splits machines into groups that cannot reach each other, though each group keeps running.
- failover — Switching work from a failed or unavailable system to a standby copy so the service keeps running.
Taught in
- Replication, Failover And Leader Election — Scalable System Design & Distributed Architecture