Replication

Replication copies data to multiple nodes for read scaling and failover. Leader-follower (primary-replica) is common: writes go to the leader, reads may hit replicas with replication lag. Multi-leader and leaderless models trade write availability for conflict resolution complexity.

Measure replication lag — stale reads break invariants for balances and permissions. Automate failover with consensus (Raft, Paxos) or orchestrated promotion with fencing to avoid split-brain writes.

On interviews: diagram primary-replica read path, state lag tolerance for your use case, and failover steps.

Common pitfalls: reading from replica immediately after write; no fencing during failover; treating replicas as backup without restore drills.

The trade-off is flexibility versus complexity—know when the simpler path is enough.

Checklist:

• Choose replication topology for consistency needs.
• Monitor and alert on replication lag.
• Define read routing rules per query type.
• Test failover and split-brain prevention.