Skip to content
master
Switch branches/tags
Go to file
Code

Varnish on Kubernetes

GitHub Workflow Status

This repository contains a controller that allows you to operate a Varnish cache on Kubernetes.


⚠️ COMPATIBILITY NOTICE: As of version v0.3, the image tag name of this project was renamed from quay.io/spaces/kube-httpcache to quay.io/mittwald/kube-httpcache. The old image will remain available (for the time being), but only the new image name will receive any updates. Please remember to adjust the image name when upgrading.


Table of Contents

How it works

This controller is not intended to be a replacement of a regular ingress controller. Instead, it is intended to be used between your regular Ingress controller and your application's service.

┌─────────┐      ┌─────────┐      ┌─────────────┐
│ Ingress ├─────▶│ Varnish ├─────▶│ Application │
└─────────┘      └─────────┘      └─────────────┘

The Varnish controller needs the following prerequisites to run:

  • A Go-template that will be used to generate a VCL configuration file
  • An application Kubernetes service that will be used as backend for the Varnish controller
  • A Varnish Kubernetes service that will be used as frontend for the Varnish controller
  • If RBAC is enabled in your cluster, you'll need a ServiceAccount with a role that grants WATCH access to the endpoints resource in the respective namespace

After starting, the Varnish controller will watch the configured Varnish service's endpoints and application service's endpoints; on startup and whenever these change, it will use the supplied VCL template to generate a new Varnish configuration and load this configuration at runtime.

The controller does not ship with any preconfigured configuration; the upstream connection and advanced features like load balancing are possible, but need to be configured in the VCL template supplied by you.

High-Availability mode

It can run in high avalability mode using multiple Varnish and application pods.

             ┌─────────┐
             │ Ingress │
             └────┬────┘
                  |
             ┌────┴────┐
             │ Service │
             └───┬┬────┘
             ┌───┘└───┐
┌────────────┴──┐  ┌──┴────────────┐
│   Varnish 1   ├──┤   Varnish 2   │
│  Signaller 1  ├──┤  Signaller 2  │
└─────────┬┬────┘  └────┬┬─────────┘
          │└─────┌──────┘│
          │┌─────┘└─────┐│
┌─────────┴┴────┐  ┌────┴┴─────────┐
│ Application 1 │  | Application 2 │
└───────────────┘  └───────────────┘

The Signaller component supports broadcasting PURGE and BAN requests to all Varnish nodes.

Getting started

Create a VCL template


⚠️ NOTE: The current implementation (supplying a VCL template as ConfigMap) may still be subject to change. Future implementations might for example use a Kubernetes Custom Resource for the entire configuration set.


Start by creating a ConfigMap that contains a VCL template:

apiVersion: v1
kind: ConfigMap
metadata:
  name: vcl-template
data:
  default.vcl.tmpl: |
    vcl 4.0;

    import std;
    import directors;

    // ".Frontends" is a slice that contains all known Varnish instances
    // (as selected by the service specified by -frontend-service).
    // The backend name needs to be the Pod name, since this value is compared
    // to the server identity ("server.identity" [1]) later.
    //
    //   [1]: https://varnish-cache.org/docs/6.4/reference/vcl.html#local-server-remote-and-client
    {{ range .Frontends }}
    backend {{ .Name }} {
        .host = "{{ .Host }}";
        .port = "{{ .Port }}";
    }
    {{- end }}

    backend fe-primary {
        .host = "{{ .PrimaryFrontend.Host }}";
        .port = "{{ .PrimaryFrontend.Port }}";
    }

    {{ range .Backends }}
    backend be-{{ .Name }} {
        .host = "{{ .Host }}";
        .port = "{{ .Port }}";
    }
    {{- end }}

    backend be-primary {
        .host = "{{ .PrimaryBackend.Host }}";
        .port = "{{ .PrimaryBackend.Port }}";
    }

    acl purgers {
        "127.0.0.1";
        "localhost";
        "::1";
        {{- range .Frontends }}
        "{{ .Host }}";
        {{- end }}
        {{- range .Backends }}
        "{{ .Host }}";
        {{- end }}
    }

    sub vcl_init {
        new cluster = directors.hash();

        {{ range .Frontends -}}
        cluster.add_backend({{ .Name }}, 1);
        {{ end }}

        new lb = directors.round_robin();

        {{ range .Backends -}}
        lb.add_backend(be-{{ .Name }});
        {{ end }}
    }

    sub vcl_recv
    {
        # Set backend hint for non cachable objects.
        set req.backend_hint = lb.backend();

        # ...

        # Routing logic. Pass a request to an appropriate Varnish node.
        # See https://info.varnish-software.com/blog/creating-self-routing-varnish-cluster for more info.
        unset req.http.x-cache;
        set req.backend_hint = cluster.backend(req.url);
        set req.http.x-shard = req.backend_hint;
        if (req.http.x-shard != server.identity) {
            return(pass);
        }
        set req.backend_hint = lb.backend();

        # ...

        return(hash);
    }

    # ...

Environment variables can be used from the template. {{ .Env.ENVVAR }} is replaced with the environment variable value. This can be used to set for example the Host-header for the external service.

Create a Secret

Create a Secret object that contains the secret for the Varnish administration port:

$ kubectl create secret generic varnish-secret --from-literal=secret=$(head -c32 /dev/urandom  | base64)

[Optional] Configure RBAC roles

If RBAC is enabled in your cluster, you will need to create a ServiceAccount with a respective Role.

$ kubectl create serviceaccount kube-httpcache
$ kubectl apply -f https://raw.githubusercontent.com/mittwald/kube-httpcache/master/deploy/kubernetes/rbac.yaml
$ kubectl create rolebinding kube-httpcache --clusterrole=kube-httpcache --serviceaccount=kube-httpcache

Deploy Varnish

  1. Create a StatefulSet for the Varnish controller:

    apiVersion: apps/v1
    kind: StatefulSet
    metadata:
      name: cache-statefulset
      labels:
        app: cache
    spec:
      serviceName: cache-service
      replicas: 2
      updateStrategy:
        type: RollingUpdate
      selector:
        matchLabels:
          app: cache
      template:
        metadata:
          labels:
            app: cache
        spec:
          containers:
          - name: cache
            image: quay.io/mittwald/kube-httpcache:stable
            imagePullPolicy: Always
            args:
            - -admin-addr=0.0.0.0
            - -admin-port=6083
            - -signaller-enable
            - -signaller-port=8090
            - -frontend-watch
            - -frontend-namespace=$(NAMESPACE)
            - -frontend-service=frontend-service
            - -backend-watch
            - -backend-namespace=$(NAMESPACE)
            - -backend-service=backend-service
            - -varnish-secret-file=/etc/varnish/k8s-secret/secret
            - -varnish-vcl-template=/etc/varnish/tmpl/default.vcl.tmpl
            - -varnish-storage=malloc,128M
            env:
            - name: NAMESPACE
              valueFrom:
                fieldRef:
                  fieldPath: metadata.namespace
            volumeMounts:
            - name: template
              mountPath: /etc/varnish/tmpl
            - name: secret
              mountPath: /etc/varnish/k8s-secret
          serviceAccountName: kube-httpcache  # when using RBAC
          restartPolicy: Always
          volumes:
          - name: template
            configMap:
              name: vcl-template
          - name: secret
            secret:
              secretName: varnish-secret

    NOTE: Using a StatefulSet is particularly important when using a stateful, self-routed Varnish cluster. Otherwise, you could also use a Deployment resource, instead.

  2. Create a service for the Varnish controller:

    apiVersion: v1
    kind: Service
    metadata:
      name: cache-service
      labels:
        app: cache
    spec:
      ports:
      - name: "http"
        port: 80
        targetPort: 80
      - name: "signaller"
        port: 8090
        targetPort: 8090
      selector:
        app: cache
  3. Create an Ingress to forward requests to cache service. Typically, you should only need an Ingress for the Services http port, and not for the signaller port (if for some reason you do, make sure to implement proper access controls)

Detailed how-tos

Using built in signaller component

The signaller component is responsible for broadcasting HTTP requests to all nodes of a Varnish cluster. This is useful in HA cluster setups, when BAN or PURGE requests should be broadcast across the entire cluster.

To broadcast a BAN or PURGE request to all Varnish endpoints, run one of the following commands, respectively:

$ curl -H "X-Url: /path" -X BAN http://cache-service:8090
$ curl -H "X-Host: www.example.com" -X PURGE http://cache-service:8090/path

When running from outside the cluster, you can use kubectl port-forward to forward the signaller port to your local machine (and then send your requests to http://localhost:8090):

$ kubectl port-forward service/cache-service 8090:8090

NOTE: Specific headers for PURGE/BAN requests depend on your Varnish configuration. E.g. X-Host header is set for convenience, because signaller is listening on other URL than Varnish. However, you need to support such headers in your VCL.

sub vcl_recv {
  # ...

  # Purge logic
  if (req.method == "PURGE") {
    if (client.ip !~ privileged) {
      return (synth(403, "Not allowed."));
    }
    if (req.http.X-Host) {
      set req.http.host = req.http.X-Host;
    }
    return (purge);
  }

  # Ban logic
  if (req.method == "BAN") {
    if (client.ip !~ privileged) {
      return (synth(403, "Not allowed."));
    }
    if (req.http.Cache-Tags) {
      ban("obj.http.Cache-Tags ~ " + req.http.Cache-Tags);
      return (synth(200, "Ban added " + req.http.host));
    }
    if (req.http.X-Url) {
      ban("obj.http.X-Url == " + req.http.X-Url);
      return (synth(200, "Ban added " + req.http.host));
    }
    return (synth(403, "Cache-Tags or X-Url header missing."));
  }

  # ...
}

Proxying to external services


NOTE: Native support for ExternalName services is a requested feature. Have a look at #39 if you're willing to help out.


In some cases, you might want to cache content from a cluster-external resource. In this case, create a new Kubernetes service of type ExternalName for your backend:

apiVersion: v1
kind: Service
metadata:
  name: external-service
  namespace: default
spec:
  type: ExternalName
  externalName: external-service.example

In your VCL template, you can then simply use this service as static backend (since there are no dynamic endpoints, you do not need to iterate over .Backends in your VCL template):

kind: ConfigMap
apiVersion: v1
metadata: # [...]
data:
  default.vcl.tmpl: |
    vcl 4.0;

    {{ range .Frontends }}
    backend {{ .Name }} {
        .host = "{{ .Host }}";
        .port = "{{ .Port }}";
    }
    {{- end }}

    backend backend {
        .host = "external-service.svc";
    }

    // ...

When starting kube-httpcache, remember to set the --backend-watch=false flag to disable watching the (non-existent) backend endpoints.

Helm Chart installation

You can use the Helm chart to rollout an instance of kube-httpcache:

$ helm install -f your-values.yaml kube-httpcache ./chart

For possible values, have a look at the comments in the provided values.yaml file.

Ensure your defined backend services have a port named http:

apiVersion: v1
kind: Service
metadata:
  name: backend-service
spec:
  ports:
  - name: http
    port: 80
    protocol: TCP
    targetPort: 8080
  type: ClusterIP

An ingress points to the kube-httpcache service which cached your backend service:

apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: example-ingress
spec:
  rules:
  - host: www.example.com
    http:
      paths:
      - backend:
          service:
            name: kube-httpcache
            port:
              number: 80
        path: /
        pathType: Prefix

Look at the vclTemplate property in chart/values.yaml to define your own Varnish cluster rules or load with extraVolume an extra file as initContainer if your ruleset is really big.