Knative Serving Architecture

Knative Serving Architecture An architecture diagram generated by Archify. Client Request · HTTP · Architecture component Client Request HTTP Gateway · Kourier / Istio · Ingress Layer Gateway Kourier / Istio User Container · Revision v1 · Data plane › Pod (Revision v1) User Container Revision v1 User Container · Revision v2 · Data plane › Pod (Revision v2) User Container Revision v2 Queue Proxy · Sidecar · Data plane › Pod (Revision v1) Queue Proxy Sidecar Queue Proxy · Sidecar · Data plane › Pod (Revision v2) Queue Proxy Sidecar Activator · Request buffering · knative-serving control plane Activator Request buffering Autoscaler · KPA / HPA · knative-serving control plane Autoscaler KPA / HPA Controller · Resource lifecycle · knative-serving control plane Controller Resource lifecycle Webhook · Validate / default · knative-serving control plane Webhook Validate / default Knative Resources · Service → Configuration → Revision / Route · Architecture component Knative Resources Service → Configuration → Revision / Route Scale > 0 Scale = 0 Forward after activation Request scale-up Concurrency metrics Scale decision Manage Validate Ingress Layer knative-serving control plane Data plane Pod (Revision v1) Pod (Revision v2) Legend Knative components / pods Knative resources (CRDs) Ingress External client

Request path

  • • With traffic, the Gateway forwards directly to Queue Proxy → User Container
  • • At scale-to-zero the first request is buffered by the Activator and forwarded once a pod is ready

Autoscaling

  • • The Queue Proxy sidecar queues requests, enforces containerConcurrency, and reports real-time concurrency
  • • KPA scales on concurrency/RPS and supports scale-to-zero; HPA scales on CPU/memory and does not

Control plane

  • • The Controller reconciles Service, Configuration, Revision, Route into Deployments, Services, and Ingress objects
  • • The Webhook validates and defaults Knative resource specs on admission