gRPC load balancing and Service Mesh
summary
Compares client-side gRPC policies to gRPCLB lookaside load balancing and proxy-based balancing, demonstrating XDS-driven Envoy sidecars (Istio or Google Traffic Director) that receive health and load reports and push weighted, locality-aware endpoints to clients. Builds a four-backend global demo (two regions, two instances each), shows live failover where clients immediately shift across continents when health changes, provides scripts that set this up in Envoy sidecar via XDS in seven steps, and notes long-lived connections require centralized capacity reporting.
AI-written summary · about
chapters (11)
- 0:00Introduction to gRPC LB
- 1:14Why load balance at scale
- 2:46Client-side load balancing
- 4:34Centralized gRPC LB
- 6:40Proxy-based load balancing
- 8:08Informed load balancing
- 11:40Locality in load balancing
- 13:53Global architecture design
- 16:03Service mesh integration
- 18:27Demonstration and setup
- 27:17Q&A and conclusion
related talks
35mgRPC Load Balancing on Kubernetes - Jan Tattermusch, Google
29mThe Security Evolution of GRPC Services in the Mesh
18mHigh Performance Service Meshes with gRPC Proxyless
25mAdapting Our XDS Control Plane for Proxyless gRPC
01h 01mgRPC client-side load balancing — why, what, how and WTF?
22mLoad Balancing in gRPC
14mEnabling Multi-Cluster Load Balancing with Gateway API and Envoy Gateway
25mBuilding Low Latency Topologies with Envoy