Definition
A read replica is a secondary copy of a database that receives a stream of changes from the primary and serves read-only queries. It is the standard first response to a read-heavy workload, because reads usually outnumber writes by a wide margin and replicas can be added without changing the data model.
What a replica does not solve is write capacity — every write still lands on one primary, so replicas postpone rather than remove the need for sharding. The cost of the arrangement is replication lag: a replica reflects the primary as of some moment in the recent past, so a user who writes and immediately reads may not see their own change. The usual fixes are routing reads to the primary for a short window after a write, pinning a session to the primary, or waiting for the replica to catch up to a known position.
How it comes up in interviews
Adding replicas is the expected first move for read scaling, and the expected follow-up is "what breaks?" Answer with the read-your-writes anomaly, describe the routing rule you would use, and be clear that replicas do nothing for write throughput — that is where sharding enters the conversation.