Serverless on Kubernetes
AWS Lambda (2014) demonstrates the serverless model: functions that scale from 0 to N automatically, billed per invocation. Many companies want the same abstraction on Kubernetes, avoiding vendor lock-in and bringing the pattern on-prem.
The release
Knative is released on 24 July 2018 by Google, with initial contributions from IBM, Red Hat, Pivotal, SAP. Apache 2.0 licence. Written in Go. Donated to CNCF on 2 March 2022 as incubating, graduated on 11 September 2025 (public announcement 8 October 2025).
Components
Knative Serving
- Scale-to-zero — pods removed when idle, restart on-demand
- Rapid scaling — scales to growing traffic
- Revisions — deployment versioning with traffic split (blue/green, canary)
- Routing — Istio/Kourier/Contour as gateway
Knative Eventing
- Sources — Kafka, GitHub webhooks, GCP Pub/Sub, Cron, AWS SQS
- Channels & Brokers — message routing
- Triggers — filters on CloudEvent schema
- CloudEvents — CNCF standard for events
apiVersion: serving.knative.dev/v1
kind: Service
metadata: { name: hello }
spec:
template:
spec:
containers:
- image: gcr.io/knative-samples/helloworld-go
env:
- name: TARGET
value: "Knative"Use cases
- Low-traffic APIs — minimal infra cost with scale-to-zero
- Event-driven workflows — reaction to webhook, queue, cron
- Function as a Service (FaaS) on K8s
- Microfrontends SSR to traffic bursts
- Batch processing triggered by events
Ecosystem and forks
- Google Cloud Run — managed Knative
- OpenFaaS — simpler OSS alternative
- Fission, Kubeless (deprecated) — predecessors
- CNCF Serverless Landscape — other alternatives
- Red Hat OpenShift Serverless — enterprise Knative distribution
Limits
- Cold start — latency on first request on scale-from-zero
- Complexity — requires Istio or alternative gateway
- Operational overhead — non-trivial to manage
- Not universal — doesn’t replace all K8s workloads
In the Italian context
Knative is used in mature Italian cloud-native teams:
- Fintech and challenger banks with burst traffic
- Large Italian e-commerce for event processing
- Digital PA — PSN projects with event-driven architectures
- AI/ML inference — model serving with scale-to-zero
- B2B SaaS — per-tenant functions
Often combined with Red Hat OpenShift or Google Cloud Run managed.
References: Knative (24 July 2018). Google + IBM + Red Hat + Pivotal + SAP. Apache 2.0 licence. Written in Go. CNCF incubating (2 March 2022), graduated (11 September 2025, announcement 8 October 2025). Serving (scale-to-zero) and Eventing (CloudEvents) components.