How do you implement custom gRPC load balancing strategies?

gRPC's load balancing is pluggable via the balancer.Builder interface. Custom load balancers implement: (1) Resolver — discovers backend addresses (from DNS, Consul, etcd, or custom service registry) and updates the balancer with address changes; (2) Balancer — receives address updates and RPC requests, selects a backend connection, and creates the SubConn; (3) Picker — makes the per-RPC selection decision. Built-in strategies: round_robin (distributes evenly), pick_first (uses first available). Custom strategies: least-loaded (track in-flight requests per connection, pick minimum); weighted round-robin (weight backends by capacity); locality-aware (prefer same-zone backends for lower latency); session affinity (route specific users to specific backends for cache warming). The xDS-based load balancing (via Envoy or Traffic Director) provides dynamic policy updates without code deployment, making it the preferred approach for advanced scenarios in Kubernetes environments.