SQL Traffic Routing Without Breaking Consistency

Read/write splitting sounds simple: send writes to the primary and reads to replicas. In production, it quickly becomes more complicated. Sessions, transactions, prepared statements, replication lag, connection pooling, retries, and failover can all affect correctness.

This talk looks at SQL traffic routing from the application’s point of view. We will discuss why routing database traffic is not the same as routing HTTP traffic, what can go wrong when replicas lag behind, when connections need to become sticky, and what signals operators should observe before making automated decisions.

The goal is to give developers, DBAs, SREs, and platform engineers a practical mental model for building safer database architectures with relational databases such as MySQL and PostgreSQL.