OpenFaaS — Serverless Functions Made Simple

What it is

OpenFaaS is a framework for packaging code, binaries, or containers as portable serverless functions. It runs on Kubernetes or containerd, providing autoscaling, metrics collection, and a rich CLI for managing deployments.

OpenFaaS appeals to developers and platform engineers who want serverless capabilities without vendor lock-in to AWS Lambda or Google Cloud Functions. If you run your own Kubernetes cluster, OpenFaaS gives you the same deploy-and-forget experience.

How it saves time or tokens

OpenFaaS eliminates boilerplate container orchestration. You write a function in any language, run faas-cli build && faas-cli deploy, and the platform handles image building, scaling to zero, and request routing. No Dockerfiles to maintain, no ingress rules to configure, no HPA manifests to write.

For AI workloads, you can wrap an inference endpoint as an OpenFaaS function and get automatic scale-to-zero when idle, cutting cloud costs on GPU instances that sit unused.

How to use

1. Install the OpenFaaS CLI: curl -sL https://cli.openfaas.com | sudo sh
2. Deploy the OpenFaaS stack to your cluster: arkade install openfaas
3. Create a new function from a template: faas-cli new --lang python3 my-function
4. Build and deploy: faas-cli up -f my-function.yml

Example

# my-function.yml
provider:
  name: openfaas
  gateway: http://127.0.0.1:8080

functions:
  my-function:
    lang: python3
    handler: ./my-function
    image: my-registry/my-function:latest
    environment:
      write_timeout: 30s
      read_timeout: 30s
    labels:
      com.openfaas.scale.min: '1'
      com.openfaas.scale.max: '10'

The function handler is a simple Python file:

def handle(req):
    return 'Hello from OpenFaaS: ' + req

Related on TokRepo

• DevOps tools -- Explore more DevOps and infrastructure automation assets
• Self-hosted tools -- Browse self-hosted alternatives to cloud services

Common pitfalls

• Forgetting to set write_timeout for long-running functions causes 502 errors under load
• Scale-to-zero is not enabled by default; you need to configure the faas-idler or use OpenFaaS Pro
• Using the community edition in production without monitoring leaves you blind to cold-start latency spikes