What are the main data replication strategies in distributed systems?

Question

Accepted Answer

Data replication maintains copies of data on multiple nodes to improve availability, durability, and read performance. Different strategies offer different trade-offs between consistency, latency, and fault tolerance.

Single-leader replication, also called master-slave or primary-secondary, designates one node as the leader that handles all writes. Followers replicate from the leader and can serve reads. This is simple to understand and implement. Writes are consistent because they go through one node. However, the leader is a single point of failure for writes, and replication lag can cause followers to serve stale data.

Multi-leader replication allows multiple nodes to accept writes. Each leader replicates to others. This improves write availability and reduces latency for geographically distributed systems. However, conflicts can occur when different leaders modify the same data concurrently. Conflict resolution strategies include last-writer-wins, which risks data loss, or application-level merging.

Leaderless replication has no designated leaders. Any node can accept reads and writes. Clients send writes to multiple nodes and read from multiple nodes. Quorum-based systems use this approach, requiring reads and writes to contact a majority of nodes. Amazon's DynamoDB and Apache Cassandra use leaderless replication. It provides high availability but requires careful handling of concurrent writes.

Synchronous replication waits for acknowledgment from replicas before confirming a write. This guarantees durability and consistency but increases latency. If a replica is slow or unavailable, writes are blocked.

Asynchronous replication confirms writes immediately after the primary writes locally, then replicates in the background. This provides lower latency but risks data loss if the primary fails before replication completes.

Most systems use semi-synchronous replication, synchronously replicating to one or a few replicas for durability while asynchronously replicating to others for performance. PostgreSQL's synchronous commit and MySQL's semi-sync replication work this way.