Understanding Knative: A Cloud-Native Serverless Framework

This article provides a comprehensive introduction to Knative, a cloud-native serverless framework built on Kubernetes. The content covers fundamental concepts including cloud-native architecture, container technology (using Linux Namespace for isolation, Cgroups for resource limits, and rootfs for filesystem), and Kubernetes basics.

Knative, open-sourced by Google in 2018 and now a CNCF incubating project, consists of three main components: Build (source-to-container building), Eventing (event-driven architecture), and Serving (traffic management and auto-scaling). The Serving component uses a sophisticated auto-scaling mechanism called KPA (Knative Pod Autoscaler) that can scale pods to zero and rapidly scale based on concurrent requests, with each pod handling up to 100 requests by default.

The article details the performance characteristics of Knative, showing benchmarks where pure pod deployment achieves 100% performance, single pod with queue-proxy achieves 85%, and multi-pod deployment achieves 40% compared to traditional HPA-based deployments. The performance overhead comes from Route, Activator, queue-proxy, and Autoscaler components.

Despite performance trade-offs, Knative offers significant advantages: improved resource utilization (25% savings in typical scenarios), rapid response to traffic spikes, simplified project maintenance through function decoupling, and enhanced developer productivity by abstracting away infrastructure concerns.