gRPC Load Balancing may not be that easy

grpc-load-balancing.md

Why gRPC load balancing is so tricky?

At a high-level, we need to understand two points:

• gRPC is built on HTTP/2, and HTTP/2 is designed to have a single long-lived TCP connection (a sticky and persistent connection);
• To do gRPC load balancing, we need to shift from connection balancing to request balancing;

Some interesting articles about this subject

• Load Balancing gRPC services
• Official Doc: gRPC Load Balancing
• gRPC Load Balancing on Kubernetes without Tears (nice explanation on tradeoffs)
• Why load balancing gRPC is tricky? (very good article)
• Old-Official Doc: Load Balancing in gRPC
• Challenges of running gRPC services in production
• Microsoft Doc: Load balancing gRPC
• Load balancing gRPC service with Nginx
• gRPC load balancing — Service Meshes
• gRPC Load Balancing inside Kubernetes
• Proxyless gRPC load balancing in Kubernetes (using xDS API)
• gRPC Loadbalancing in GKE using Nginx Ingress Controller
• On gRPC Load Balancing

Google Cloud has invested in Proxyless gRPC services (using xDS API)

• Traffic Director and gRPC—proxyless services for your service mesh
• Traffic Director with proxyless gRPC services overview
• Proxyless gRPC adds support for advanced traffic management features (timeouts, circuit breakers and fault-injection tests)

A little bit about HTTP/2

https://www.cncf.io/blog/2018/07/03/http-2-smarter-at-scale/

https://grpc.io/blog/grpc-on-http2/

https://www.lucidchart.com/techblog/2019/04/10/why-turning-on-http2-was-a-mistake/ - um pouco superficial sem muitos detalhes

I'm working on this with some advances.
https://gitlab.com/infinito84/grpc-go-kubernetes

From inside the cluster, with linkerd I can handle load balancing at L7, however, I haven't been able to use this load balancing feature outside the cluster. could you give a hand?