One of Kubernetes' most basic functions is load balancing: routing traffic among several pods so that, ideally, each pod handles the same amount of load. But, Kubernetes by itself, only balances connections rather than requests, and that has some downsides for protocols like gRPC and HTTP/2.

The good news: a service mesh can do better. Since service meshes mediate all application traffic, they can do per-request load balancing for gRPC and HTTP/2, making for much more effective resource utilization than per-connection balancing. Since they understand what happens in each request, they can also provide reliability and observability at the request level (for example, retrying only failed gRPC requests). And they can provide all this additional functionality without needing to change the application at all. Join Dominik for a deep dive into what service mesh can do, including a live demonstration using Linkerd and paying special attention to streaming gRPC.