HTTP/2 and gRPC Load Balancing
What You'll Learn
You'll learn how to configure load balancers to handle modern protocols like HTTP/2 and gRPC, which behave differently from traditional HTTP/1.1 traffic.
The Core Challenge
Traditional HTTP/1.1 creates a new connection for each request or maintains multiple connections per client. HTTP/2 and gRPC, however, use multiplexing—sending many requests simultaneously over a single, long-lived TCP connection.
Why This Matters for Load Balancing
Imagine a highway with many lanes (HTTP/1.1) versus a single super-efficient lane where cars can overlap and pass through simultaneously (HTTP/2). Traditional load balancers distribute connections across backend servers, but with HTTP/2, all requests from one client might stick to a single backend through one connection—creating uneven load distribution.
How to Handle HTTP/2 and gRPC
Application Load Balancers (ALB) automatically handle HTTP/2 by:
- Accepting HTTP/2 connections from clients
- Converting them to HTTP/1.1 for backend targets (or maintaining HTTP/2 end-to-end)
- Distributing individual requests across target groups, even within a single connection
For gRPC (which runs over HTTP/2):
- ALB supports gRPC natively with proper routing based on service names and methods
- Health checks work with gRPC-specific checks
- You must enable HTTP/2 protocol support on the load balancer listener
Network Load Balancers (NLB) handle gRPC differently:
- Pass-through at Layer 4 maintains the long-lived connection
- Useful when you need end-to-end encryption or client IP preservation
- Backends must handle load distribution themselves
Configuration Considerations
Enable HTTP/2 support on your ALB listener, ensure your target groups understand gRPC health checks, and monitor connection distribution to avoid "sticky connection" imbalances where one backend gets overloaded.
Key Takeaway: HTTP/2 and gRPC use persistent multiplexed connections that require Layer 7 (ALB) load balancing to properly distribute individual requests, not just connections, across backend servers.