05-canary.md

Canary deployments for OpenFaaS functions

Create a general available release for the env function version 1.0.0:

apiVersion: openfaas.com/v1alpha2
kind: Function
metadata:
  name: env
  namespace: openfaas-fn
spec:
  name: env
  image: stefanprodan/of-env:1.0.0
  resources:
    requests:
      memory: "32Mi"
      cpu: "10m"
  limits:
    memory: "64Mi"
    cpu: "100m"

Save the above resources as env-ga.yaml and then apply it:

kubectl apply -f ./env-ga.yaml

Create a canary release for version 1.1.0:

apiVersion: openfaas.com/v1alpha2
kind: Function
metadata:
  name: env-canary
  namespace: openfaas-fn
spec:
  name: env-canary
  image: stefanprodan/of-env:1.1.0
  resources:
    requests:
      memory: "32Mi"
      cpu: "10m"
  limits:
    memory: "64Mi"
    cpu: "100m"

Save the above resources as env-canary.yaml and then apply it:

kubectl apply -f ./env-canary.yaml

Create an Istio virtual service with 10% traffic going to canary:

apiVersion: networking.istio.io/v1alpha3
kind: VirtualService
metadata:
  name: env
  namespace: openfaas-fn
spec:
  hosts:
  - env
  http:
  - route:
    - destination:
        host: env
      weight: 90
    - destination:
        host: env-canary
      weight: 10
    timeout: 30s

Save the above resources as env-virtual-service.yaml and then apply it:

kubectl apply -f ./env-virtual-service.yaml

Test traffic routing (one in ten calls should hit the canary release):

 while true; do sleep 1; curl -sS https://openfaas.example.com/function/env | grep HOSTNAME; done

HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-59bf48fb9d-cjsjw
HOSTNAME=env-canary-5dffdf4458-4vnn2

Access Jaeger dashboard using port forwarding:

kubectl -n istio-system port-forward deployment/istio-tracing 16686:16686

Tracing the general available release:

Tracing the canary release:

Monitor ga vs canary success rate and latency with Grafana: