This guide shows you how to use the NGINX ingress controller and Flagger to automate canary deployments and A/B testing.
Flagger requires a Kubernetes cluster v1.16 or newer and NGINX ingress v0.41 or newer.
Install the NGINX ingress controller with Helm v3:
helm repo add ingress-nginx https://kubernetes.github.io/ingress-nginxkubectl create ns ingress-nginxhelm upgrade -i ingress-nginx ingress-nginx/ingress-nginx \--namespace ingress-nginx \--set controller.metrics.enabled=true \--set controller.podAnnotations."prometheus\.io/scrape"=true \--set controller.podAnnotations."prometheus\.io/port"=10254
Install Flagger and the Prometheus add-on in the same namespace as the ingress controller:
helm repo add flagger https://flagger.app​helm upgrade -i flagger flagger/flagger \--namespace ingress-nginx \--set prometheus.install=true \--set meshProvider=nginx
Flagger takes a Kubernetes deployment and optionally a horizontal pod autoscaler (HPA), then creates a series of objects (Kubernetes deployments, ClusterIP services and canary ingress). These objects expose the application outside the cluster and drive the canary analysis and promotion.
Create a test namespace:
kubectl create ns test
Create a deployment and a horizontal pod autoscaler:
kubectl apply -k https://github.com/fluxcd/flagger//kustomize/podinfo?ref=main
Deploy the load testing service to generate traffic during the canary analysis:
helm upgrade -i flagger-loadtester flagger/loadtester \--namespace=test
Create an ingress definition (replace app.example.com
with your own domain):
apiVersion: networking.k8s.io/v1beta1kind: Ingressmetadata:name: podinfonamespace: testlabels:app: podinfoannotations:kubernetes.io/ingress.class: "nginx"spec:rules:- host: app.example.comhttp:paths:- backend:serviceName: podinfoservicePort: 80
Save the above resource as podinfo-ingress.yaml and then apply it:
kubectl apply -f ./podinfo-ingress.yaml
Create a canary custom resource (replace app.example.com
with your own domain):
apiVersion: flagger.app/v1beta1kind: Canarymetadata:name: podinfonamespace: testspec:provider: nginx# deployment referencetargetRef:apiVersion: apps/v1kind: Deploymentname: podinfo# ingress referenceingressRef:apiVersion: networking.k8s.io/v1beta1kind: Ingressname: podinfo# HPA reference (optional)autoscalerRef:apiVersion: autoscaling/v2beta2kind: HorizontalPodAutoscalername: podinfo# the maximum time in seconds for the canary deployment# to make progress before it is rollback (default 600s)progressDeadlineSeconds: 60service:# ClusterIP port numberport: 80# container port number or nametargetPort: 9898analysis:# schedule interval (default 60s)interval: 10s# max number of failed metric checks before rollbackthreshold: 10# max traffic percentage routed to canary# percentage (0-100)maxWeight: 50# canary increment step# percentage (0-100)stepWeight: 5# NGINX Prometheus checksmetrics:- name: request-success-rate# minimum req success rate (non 5xx responses)# percentage (0-100)thresholdRange:min: 99interval: 1m# testing (optional)webhooks:- name: acceptance-testtype: pre-rollouturl: http://flagger-loadtester.test/timeout: 30smetadata:type: bashcmd: "curl -sd 'test' http://podinfo-canary/token | grep token"- name: load-testurl: http://flagger-loadtester.test/timeout: 5smetadata:cmd: "hey -z 1m -q 10 -c 2 http://app.example.com/"
Save the above resource as podinfo-canary.yaml and then apply it:
kubectl apply -f ./podinfo-canary.yaml
After a couple of seconds Flagger will create the canary objects:
# applieddeployment.apps/podinfohorizontalpodautoscaler.autoscaling/podinfoingresses.extensions/podinfocanary.flagger.app/podinfo​# generateddeployment.apps/podinfo-primaryhorizontalpodautoscaler.autoscaling/podinfo-primaryservice/podinfoservice/podinfo-canaryservice/podinfo-primaryingresses.extensions/podinfo-canary
Flagger implements a control loop that gradually shifts traffic to the canary while measuring key performance indicators like HTTP requests success rate, requests average duration and pod health. Based on analysis of the KPIs a canary is promoted or aborted, and the analysis result is published to Slack or MS Teams.
Trigger a canary deployment by updating the container image:
kubectl -n test set image deployment/podinfo \podinfod=stefanprodan/podinfo:3.1.1
Flagger detects that the deployment revision changed and starts a new rollout:
kubectl -n test describe canary/podinfo​Status:Canary Weight: 0Failed Checks: 0Phase: SucceededEvents:Type Reason Age From Message---- ------ ---- ---- -------Normal Synced 3m flagger New revision detected podinfo.testNormal Synced 3m flagger Scaling up podinfo.testWarning Synced 3m flagger Waiting for podinfo.test rollout to finish: 0 of 1 updated replicas are availableNormal Synced 3m flagger Advance podinfo.test canary weight 5Normal Synced 3m flagger Advance podinfo.test canary weight 10Normal Synced 3m flagger Advance podinfo.test canary weight 15Normal Synced 2m flagger Advance podinfo.test canary weight 20Normal Synced 2m flagger Advance podinfo.test canary weight 25Normal Synced 1m flagger Advance podinfo.test canary weight 30Normal Synced 1m flagger Advance podinfo.test canary weight 35Normal Synced 55s flagger Advance podinfo.test canary weight 40Normal Synced 45s flagger Advance podinfo.test canary weight 45Normal Synced 35s flagger Advance podinfo.test canary weight 50Normal Synced 25s flagger Copying podinfo.test template spec to podinfo-primary.testWarning Synced 15s flagger Waiting for podinfo-primary.test rollout to finish: 1 of 2 updated replicas are availableNormal Synced 5s flagger Promotion completed! Scaling down podinfo.test
Note that if you apply new changes to the deployment during the canary analysis, Flagger will restart the analysis.
You can monitor all canaries with:
watch kubectl get canaries --all-namespaces​NAMESPACE NAME STATUS WEIGHT LASTTRANSITIONTIMEtest podinfo Progressing 15 2019-05-06T14:05:07Zprod frontend Succeeded 0 2019-05-05T16:15:07Zprod backend Failed 0 2019-05-04T17:05:07Z
During the canary analysis you can generate HTTP 500 errors to test if Flagger pauses and rolls back the faulted version.
Trigger another canary deployment:
kubectl -n test set image deployment/podinfo \podinfod=stefanprodan/podinfo:3.1.2
Generate HTTP 500 errors:
watch curl http://app.example.com/status/500
When the number of failed checks reaches the canary analysis threshold, the traffic is routed back to the primary, the canary is scaled to zero and the rollout is marked as failed.
kubectl -n test describe canary/podinfo​Status:Canary Weight: 0Failed Checks: 10Phase: FailedEvents:Type Reason Age From Message---- ------ ---- ---- -------Normal Synced 3m flagger Starting canary deployment for podinfo.testNormal Synced 3m flagger Advance podinfo.test canary weight 5Normal Synced 3m flagger Advance podinfo.test canary weight 10Normal Synced 3m flagger Advance podinfo.test canary weight 15Normal Synced 3m flagger Halt podinfo.test advancement success rate 69.17% < 99%Normal Synced 2m flagger Halt podinfo.test advancement success rate 61.39% < 99%Normal Synced 2m flagger Halt podinfo.test advancement success rate 55.06% < 99%Normal Synced 2m flagger Halt podinfo.test advancement success rate 47.00% < 99%Normal Synced 2m flagger (combined from similar events): Halt podinfo.test advancement success rate 38.08% < 99%Warning Synced 1m flagger Rolling back podinfo.test failed checks threshold reached 10Warning Synced 1m flagger Canary failed! Scaling down podinfo.test
The canary analysis can be extended with Prometheus queries.
The demo app is instrumented with Prometheus so you can create a custom check that will use the HTTP request duration histogram to validate the canary.
Create a metric template and apply it on the cluster:
apiVersion: flagger.app/v1beta1kind: MetricTemplatemetadata:name: latencynamespace: testspec:provider:type: prometheusaddress: http://flagger-prometheus.ingress-nginx:9090query: |histogram_quantile(0.99,sum(rate(http_request_duration_seconds_bucket{kubernetes_namespace="{{ namespace }}",kubernetes_pod_name=~"{{ target }}-[0-9a-zA-Z]+(-[0-9a-zA-Z]+)"}[1m])) by (le))
Edit the canary analysis and add the latency check:
analysis:metrics:- name: "latency"templateRef:name: latencythresholdRange:max: 0.5interval: 1m
The threshold is set to 500ms so if the average request duration in the last minute goes over half a second then the analysis will fail and the canary will not be promoted.
Trigger a canary deployment by updating the container image:
kubectl -n test set image deployment/podinfo \podinfod=stefanprodan/podinfo:3.1.3
Generate high response latency:
watch curl http://app.example.com/delay/2
Watch Flagger logs:
kubectl -n nginx-ingress logs deployment/flagger -f | jq .msg​Starting canary deployment for podinfo.testAdvance podinfo.test canary weight 5Advance podinfo.test canary weight 10Advance podinfo.test canary weight 15Halt podinfo.test advancement latency 1.20 > 0.5Halt podinfo.test advancement latency 1.45 > 0.5Halt podinfo.test advancement latency 1.60 > 0.5Halt podinfo.test advancement latency 1.69 > 0.5Halt podinfo.test advancement latency 1.70 > 0.5Rolling back podinfo.test failed checks threshold reached 5Canary failed! Scaling down podinfo.test
If you have alerting configured, Flagger will send a notification with the reason why the canary failed.
Besides weighted routing, Flagger can be configured to route traffic to the canary based on HTTP match conditions. In an A/B testing scenario, you'll be using HTTP headers or cookies to target a certain segment of your users. This is particularly useful for frontend applications that require session affinity.
Edit the canary analysis, remove the max/step weight and add the match conditions and iterations:
analysis:interval: 1mthreshold: 10iterations: 10match:# curl -H 'X-Canary: insider' http://app.example.com- headers:x-canary:exact: "insider"# curl -b 'canary=always' http://app.example.com- headers:cookie:exact: "canary"metrics:- name: request-success-ratethresholdRange:min: 99interval: 1mwebhooks:- name: load-testurl: http://flagger-loadtester.test/timeout: 5smetadata:cmd: "hey -z 1m -q 10 -c 2 -H 'Cookie: canary=always' http://app.example.com/"
The above configuration will run an analysis for ten minutes targeting users that have a canary
cookie set to always
or those that call the service using the X-Canary: insider
header.
Trigger a canary deployment by updating the container image:
kubectl -n test set image deployment/podinfo \podinfod=stefanprodan/podinfo:3.1.4
Flagger detects that the deployment revision changed and starts the A/B testing:
kubectl -n test describe canary/podinfo​Status:Failed Checks: 0Phase: SucceededEvents:Type Reason Age From Message---- ------ ---- ---- -------Normal Synced 3m flagger New revision detected podinfo.testNormal Synced 3m flagger Scaling up podinfo.testWarning Synced 3m flagger Waiting for podinfo.test rollout to finish: 0 of 1 updated replicas are availableNormal Synced 3m flagger Advance podinfo.test canary iteration 1/10Normal Synced 3m flagger Advance podinfo.test canary iteration 2/10Normal Synced 3m flagger Advance podinfo.test canary iteration 3/10Normal Synced 2m flagger Advance podinfo.test canary iteration 4/10Normal Synced 2m flagger Advance podinfo.test canary iteration 5/10Normal Synced 1m flagger Advance podinfo.test canary iteration 6/10Normal Synced 1m flagger Advance podinfo.test canary iteration 7/10Normal Synced 55s flagger Advance podinfo.test canary iteration 8/10Normal Synced 45s flagger Advance podinfo.test canary iteration 9/10Normal Synced 35s flagger Advance podinfo.test canary iteration 10/10Normal Synced 25s flagger Copying podinfo.test template spec to podinfo-primary.testWarning Synced 15s flagger Waiting for podinfo-primary.test rollout to finish: 1 of 2 updated replicas are availableNormal Synced 5s flagger Promotion completed! Scaling down podinfo.test
The above procedure can be extended with custom metrics checks, webhooks, manual promotion approval and Slack or MS Teams notifications.