What is Failover?

Failover is a fault-tolerance mechanism in database systems that transfers control from a failed primary node to a standby replica. When the active database instance becomes unreachable due to hardware failure, software crash, or network partition, the system or an administrator promotes a designated replica to take over as the new primary. This transition can be automatic, triggered by a health-check timeout or a consensus protocol, or manual, requiring an operator to verify the failure and initiate the promotion.

The process typically involves several steps. First, the failure must be detected, often through heartbeat signals or a distributed consensus algorithm such as Raft or Paxos. Next, the system ensures that any in-flight transactions are either committed or rolled back, and the replica catches up to the last consistent state using a write-ahead log or replication stream. Finally, the replica is promoted, its role is updated in the cluster metadata, and client connections are redirected to the new primary. The time from failure to full read-write availability is called the recovery time objective (RTO).

Failover is a core component of high-availability database architectures, including active-passive clusters, primary-replica replication, and multi-region deployments. It differs from switchover, which is a planned, graceful role reversal. Failover alone does not guarantee data durability; synchronous replication or semi-synchronous replication is often paired with failover to minimize data loss. Common implementations include PostgreSQL streaming replication with Patroni, MySQL InnoDB Cluster with Group Replication, and MongoDB replica sets.