Read replicas are great for scaling reads, but what if your read replicas need to be elastic…
This is where you put read replicas behind a proxy like HAProxy or a classic load balancer. Clients (say, your app) connect to a single endpoint, and the proxy routes queries across all available replicas.
Here’s what makes this interesting…
The proxy can perform health checks on each replica, automatically removing failed or lagging instances from the pool. This means your reads stay fast even when individual replicas fall behind or go down.
You can add or remove replicas based on load without touching your application code. During peak hours, scale up. During quiet periods, scale down. The proxy handles all the routing complexity.
Most proxies also support weighted load balancing. If one replica has better specs or lower latency, you can send it proportionally more traffic. This optimizes resource utilization across your replica fleet.
The setup also gives you massive deployment flexibility, especially during a replica upgrade, where you can route traffic away from it, upgrade, verify, and then bring it back into rotation. Zero downtime :)
This architecture comes in handy when infrastructure needs to be as dynamic as the traffic pattern to keep costs down, and it shows up across a lot of companies and is an easy win for most products operating at scale.
Hope this helps.