Leader–follower replication

Problem

One node can't handle all reads. And if the node dies, you lose everything.

Context

You have a read-heavy workload and need horizontal read scale + failover.

Solution

One node is the leader; it accepts all writes. N followers replicate the leader's write log asynchronously. Reads can go to any node. On leader failure, promote a follower to leader.

Trade-offs

• Followers are eventually consistent — replication lag causes read-after-write anomalies
• Writes still serialized through the leader — no write scale
• Failover has an outage window (~seconds) for leader promotion
• Synchronous replication improves consistency at the cost of write latency

Failure modes

• Replication lag grows unbounded under write load — followers fall behind, reads get stale
• Split-brain during network partition — old leader and new leader both accept writes
• Follower promoted to leader but wasn't caught up — data loss

When to use

• Read-heavy workloads (100:1 or higher read:write ratio)
• You need HA with automatic failover
• Some staleness in read replicas is acceptable

When NOT to use

• Write-heavy workloads — leader is still the bottleneck
• Strong read-after-write consistency required — either read from leader or use synchronous replication