Yes, you will need to set up another Cloud SQL Proxy instance to connect to your read replica. To connect a Google Cloud SQL read replica to a Kubernetes Engine (GKE) pod using the Cloud SQL Proxy, a separate proxy instance is required. This is because each Cloud SQL Proxy is tied to a specific instance, so the read replica, being a distinct instance, necessitates its own proxy configuration.
For load balancing across multiple read replicas, the recommended approach is to use HAProxy, as Google Cloud SQL does not provide a built-in load balancer for read replicas. HAProxy can be deployed within GKE to distribute read workloads effectively. Alternative solutions include using PostgreSQL middleware tools like Pgpool-II or PgBouncer, which offer connection pooling and load balancing capabilities and can be deployed as sidecars or separate services in GKE. Additionally, client-side load balancing (if supported by the PostgreSQL client or ORM) and DNS-based round-robin can be considered, though the latter lacks health checks, making it less ideal for production.
To implement HAProxy, deploy it as a Kubernetes service within GKE, listing all read replicas as backend servers in the HAProxy configuration. This setup allows health checks and failover configurations to ensure only healthy replicas are routed traffic. Applications should then be configured to connect to the HAProxy service for read operations, ensuring efficient load distribution across replicas.
LinkedIn
Twitter
