knative-serving

Knative Serving enables serverless deployment on Kubernetes with request-based autoscaling and scale-to-zero capability. Create a Knative Service manifest specifying your container image, set autoscaling policies (concurrency-based or CPU-based), and Knative automatically scales instances up with demand and down to zero during idle periods.

Yes. Knative Serving supports traffic splitting across revisions, enabling canary deployments, blue-green rollouts, and A/B testing. Define traffic percentages in your service spec to gradually route requests to new revisions while monitoring performance before full cutover.

Knative Serving supports concurrency-based autoscaling (scale on concurrent requests), CPU-based autoscaling, and custom metrics. Configure autoscaling through annotations and containerConcurrency settings in your service manifest to match your workload's scaling behavior.

Set timeoutSeconds, containerConcurrency, and resource requests/limits in your Knative Service spec. These control request timeout behavior, concurrent request handling per container, and CPU/memory allocation, ensuring predictable performance and cost.

Readiness probes are optional but recommended for Knative Services to signal when your container is ready to receive traffic. Configure them in your pod spec to prevent traffic routing to unhealthy instances, improving reliability during scale events.

Knative Serving provides request-driven autoscaling and scale-to-zero, reducing costs during low-traffic periods. Traditional Kubernetes deployments scale based on resource metrics but maintain minimum replicas, making Knative ideal for variable-demand workloads like AI agents.