What is Read Replica?

A read replica is a secondary database instance that maintains a near-real-time copy of data from a primary (source) database. The replica is updated asynchronously, meaning the primary does not wait for the replica to receive or apply changes before completing its own write transaction. This design prevents replicas from slowing down write-heavy workloads on the primary, but it introduces a small replication lag typically measured in milliseconds to seconds.

The read replica's primary purpose is to handle read queries such as SELECT statements, reporting, analytics, or data extract jobs. By directing these queries to replicas instead of the primary, operators reduce contention on the primary's CPU, memory, and I/O resources. Read replicas are common in relational database systems (e.g., MySQL, PostgreSQL, Amazon RDS, Azure SQL Database) and are also used in some NoSQL stores (e.g., MongoDB, DynamoDB Global Tables with read-only replicas). Replication is usually log-based or stream-based: the primary streams its transaction log or change data capture (CDC) stream to the replica, which applies the changes in order.

In a cloud or distributed architecture, read replicas can be deployed in different availability zones or geographic regions for better disaster recovery and lower latency for geographically distributed users. However, because the replication is asynchronous, the replica may not reflect the latest writes at any given moment. Applications that require strong consistency should always read from the primary. Some database services allow promoting a read replica to a standalone primary in case of failure, though this may cause some data loss equal to the replication lag.