Its important to set allowEmptyServices to true, otherwhise the scale up will not work because traefik cannot find the service if it was scaled down to zero.
- "--pilot.token=xxxx"
- "--experimental.plugins.traefik-ondemand-plugin.modulename=github.com/acouvreur/traefik-ondemand-plugin"
- "--experimental.plugins.traefik-ondemand-plugin.version=v0.1.1"
- "--providers.kubernetesingress.allowEmptyServices=true"
If you are using the traefik helm chart its also important to set:
experimental:
plugins:
enabled: true
In this example we will deploy the traefik-ondemand-service into the namespace kube-system
apiVersion: apps/v1
kind: Deployment
metadata:
name: traefik-ondemand-service
namespace: kube-system
labels:
app: traefik-ondemand-service
spec:
replicas: 1
selector:
matchLabels:
app: traefik-ondemand-service
template:
metadata:
labels:
app: traefik-ondemand-service
spec:
serviceAccountName: traefik-ondemand-service
serviceAccount: traefik-ondemand-service
containers:
- name: traefik-ondemand-service
image: gchr.io/acouvreur/traefik-ondemand-service
args: ["--swarmMode=false", "--kubernetesMode=true"]
ports:
- containerPort: 10000
---
apiVersion: v1
kind: Service
metadata:
name: traefik-ondemand-service
namespace: kube-system
spec:
selector:
app: traefik-ondemand-service
ports:
- protocol: TCP
port: 10000
targetPort: 10000
We have to create RBAC to allow the traefik-ondemand-service to access the kubernetes API and get/update/patch the deployment resource
apiVersion: v1
kind: ServiceAccount
metadata:
name: traefik-ondemand-service
namespace: kube-system
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
name: traefik-ondemand-service
namespace: kube-system
rules:
- apiGroups:
- apps
resources:
- statefulsets
- statefulsets/scale
- deployments
- deployments/scale
verbs:
- patch
- get
- update
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: traefik-ondemand-service
namespace: kube-system
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: traefik-ondemand-service
subjects:
- kind: ServiceAccount
name: traefik-ondemand-service
namespace: kube-system
In this example we want to scale down the code-server
deployment in the codeserverns
namespace
First we need to create a traefik middleware for that:
apiVersion: traefik.containo.us/v1alpha1
kind: Middleware
metadata:
name: ondemand-codeserver
namespace: kube-system
spec:
plugin:
traefik-ondemand-plugin:
name: deployment_codeserverns_code-server_1
serviceUrl: 'http://traefik-ondemand-service:10000'
timeout: 10m
The format of the name:
section is <KIND>_<NAMESPACE>_<NAME>_<REPLICACOUNT>
where _
is the delimiter.
KIND
can be either deployment
or statefulset
When using an Ingress (e.g. for code-server) you have to add the middleware in metadata.annotation:
traefik.ingress.kubernetes.io/router.middlewares: kube-system-ondemand-codeserver@kubernetescrd