☰
Navigation :

Building a Proxy Controller for Gloo Edge

In this tutorial, we’re going to show how to use Gloo Edge’s Proxy API to build a router which automatically creates routes for every existing kubernetes service. Then we will show how to enable the same functionality for consul services as well.

Why Write a Custom Proxy Controller

Building a Proxy controller allows you to add custom Gloo Edge operational logic to your setup. In this example, we will write a proxy controller that creates and manages a second Gloo Edge Proxy (my-cool-proxy) alongside the Gloo Edge managed one (gateway-proxy). The my-cool-proxy envoy proxy will route to Gloo Edge discovered kubernetes services using the host header alone, relieving the Gloo Edge admin from creating virtual services or route tables to route to each discovered service.

Other common use cases that can be solved with custom proxy controllers include:

How it will work

A custom Proxy controller takes any inputs (in our case, Gloo Edge custom resources in kubernetes) and writes the desired output to managed Proxy custom resource(s).

In our case, we will write a controller that takes Upstreams and Proxys as inputs and outputs a new Proxy. Then we will deploy the new controller to create and manage our new my-cool-proxy Proxy custom resource. Finally, we will deploy a second envoy proxy to kubernetes, have it register to Gloo Edge with its role configured to match the name of our managed Proxy custom resource (my-cool-proxy), and configure it to receive configuration from Gloo Edge.

Writing the Code

Tested with Gloo Edge v1.2.12 and Gloo Edge Enterprise v1.2.0.

You can view the complete code written in this section here.

If you’d prefer, you can skip to running the example and tinker with that as desired rather than follow along below.

Dependencies

tested with go version 1.13.5

First, initialize a new go module. This should be done in a new directory. The command to initialize an empty go module is:

go mod init <your module name here>

Once the module has been initialized add the following dependencies to your go.mod file. Kubernetes dependency management with go.mod can be fairly difficult, this should be all you need to get access to their types, as well as ours.

Click to see the full go.mod file that should be used for this project

The basis of this go.mod file is from the Gloo Edge go.mod file

Now that the dependencies are complete (for now), we can move on to the interesting part: writing the controller!

Initial code

First, we’ll start with a main.go. We’ll use the main function to connect to Kubernetes and start an event loop. Start by creating a new main.go file in a new directory:

package main

// all the import's we'll need for this controller
import (
	"context"
	"log"
	"os"
	"time"

	v1 "github.com/solo-io/gloo/projects/gloo/pkg/api/v1"
	matchers "github.com/solo-io/gloo/projects/gloo/pkg/api/v1/core/matchers"
	"github.com/solo-io/go-utils/kubeutils"
	"github.com/solo-io/solo-kit/pkg/api/v1/clients"
	"github.com/solo-io/solo-kit/pkg/api/v1/clients/factory"
	"github.com/solo-io/solo-kit/pkg/api/v1/clients/kube"
	core "github.com/solo-io/solo-kit/pkg/api/v1/resources/core"

	// import for GKE
	_ "k8s.io/client-go/plugin/pkg/client/auth/gcp"
)


func main() {}


// make our lives easy
func must(err error) {
	if err != nil {
		panic(err)
	}
}

Gloo Edge API Clients

Then we’ll want to use Gloo Edge’s libraries to initialize a client for Proxies and Upstreams. Add the following function to your code:


func initGlooClients(ctx context.Context) (v1.UpstreamClient, v1.ProxyClient) {
	// root rest config
	restConfig, err := kubeutils.GetConfig(
		os.Getenv("KUBERNETES_MASTER_URL"),
		os.Getenv("KUBECONFIG"))
	must(err)

	// wrapper for kubernetes shared informer factory
	cache := kube.NewKubeCache(ctx)

	// initialize the CRD client for Gloo Edge Upstreams
	upstreamClient, err := v1.NewUpstreamClient(&factory.KubeResourceClientFactory{
		Crd:         v1.UpstreamCrd,
		Cfg:         restConfig,
		SharedCache: cache,
	})
	must(err)

	// registering the client registers the type with the client cache
	err = upstreamClient.Register()
	must(err)

	// initialize the CRD client for Gloo Edge Proxies
	proxyClient, err := v1.NewProxyClient(&factory.KubeResourceClientFactory{
		Crd:         v1.ProxyCrd,
		Cfg:         restConfig,
		SharedCache: cache,
	})
	must(err)

	// registering the client registers the type with the client cache
	err = proxyClient.Register()
	must(err)

	return upstreamClient, proxyClient
}

This function will initialize clients for interacting with Gloo Edge’s Upstream and Proxy APIs.

Proxy Configuration

Next, we’ll define the algorithm for generating Proxy CRDs from a given list of upstreams. In this example, our proxy will serve traffic to every service in our cluster.

Paste the following function into your code. Feel free to modify if you want to get experimental, here’s where the “opinionated” piece of our controller is defined:


// in this function we'll generate an opinionated
// proxy object with a routes for each of our upstreams
func makeDesiredProxy(upstreams v1.UpstreamList) *v1.Proxy {

	// each virtual host represents the table of routes for a given
	// domain or set of domains.
	// in this example, we'll create one virtual host
	// for each upstream.
	var virtualHosts []*v1.VirtualHost

	for _, upstream := range upstreams {
		upstreamRef := upstream.Metadata.Ref()
		// create a virtual host for each upstream
		vHostForUpstream := &v1.VirtualHost{
			// logical name of the virtual host, should be unique across vhosts
			Name: upstream.Metadata.Name,

			// the domain will be our "matcher".
			// requests with the Host header equal to the upstream name
			// will be routed to this upstream
			Domains: []string{upstream.Metadata.Name},

			// we'll create just one route designed to match any request
			// and send it to the upstream for this domain
			Routes: []*v1.Route{{
				// use a basic catch-all matcher
				Matchers: []*matchers.Matcher{
					&matchers.Matcher{
						PathSpecifier: &matchers.Matcher_Prefix{
							Prefix: "/",
						},
					},
				},

				// tell Gloo Edge where to send the requests
				Action: &v1.Route_RouteAction{
					RouteAction: &v1.RouteAction{
						Destination: &v1.RouteAction_Single{
							// single destination
							Single: &v1.Destination{
								DestinationType: &v1.Destination_Upstream{
									// a "reference" to the upstream, which is a Namespace/Name tuple
									Upstream: &upstreamRef,
								},
							},
						},
					},
				},
			}},
		}

		virtualHosts = append(virtualHosts, vHostForUpstream)
	}

	desiredProxy := &v1.Proxy{
		// metadata will be translated to Kubernetes ObjectMeta
		Metadata: core.Metadata{Namespace: "gloo-system", Name: "my-cool-proxy"},

		// we have the option of creating multiple listeners,
		// but for the purpose of this example we'll just use one
		Listeners: []*v1.Listener{{
			// logical name for the listener
			Name: "my-amazing-listener",

			// instruct envoy to bind to all interfaces on port 8080
			BindAddress: "::", BindPort: 8080,

			// at this point you determine what type of listener
			// to use. here we'll be using the HTTP Listener
			// other listener types are currently unsupported,
			// but future
			ListenerType: &v1.Listener_HttpListener{
				HttpListener: &v1.HttpListener{
					// insert our list of virtual hosts here
					VirtualHosts: virtualHosts,
				},
			}},
		},
	}

	return desiredProxy
}

Event Loop

Now we’ll define a resync function to be called whenever we receive a new list of upstreams:


// we received a new list of upstreams! regenerate the desired proxy
// and write it as a CRD to Kubernetes
func resync(ctx context.Context, upstreams v1.UpstreamList, client v1.ProxyClient) {
	desiredProxy := makeDesiredProxy(upstreams)

	// see if the proxy exists. if yes, update; if no, create
	existingProxy, err := client.Read(
		desiredProxy.Metadata.Namespace,
		desiredProxy.Metadata.Name,
		clients.ReadOpts{Ctx: ctx})


	// proxy exists! this is an update, not a create
	if err == nil {

		// sleep for 1s as Gloo Edge may be re-validating our proxy, which can cause resource version to change
		time.Sleep(time.Second)

		// ensure resource version is the latest
		existingProxy, err = client.Read(
			desiredProxy.Metadata.Namespace,
			desiredProxy.Metadata.Name,
			clients.ReadOpts{Ctx: ctx})
		must(err)

		// update the resource version on our desired proxy
		desiredProxy.Metadata.ResourceVersion = existingProxy.Metadata.ResourceVersion
	}

	// write!
	written, err := client.Write(desiredProxy,
		clients.WriteOpts{Ctx: ctx, OverwriteExisting: true})

	must(err)

	log.Printf("wrote proxy object: %+v\n", written)
}

Main Function

Now that we have our clients and a function defining the proxies we’ll want to create, all we need to do is tie it all together.

Let’s set up a loop to watch Upstreams in our main function. Add the following to your main() func:


func main() {
	// root context for the whole thing
	ctx := context.Background()

	// initialize Gloo Edge API clients
	upstreamClient, proxyClient := initGlooClients(ctx)

	// start a watch on upstreams. we'll use this as our trigger
	// whenever upstreams are modified, we'll trigger our sync function
	upstreamWatch, watchErrors, initError := upstreamClient.Watch("gloo-system",
		clients.WatchOpts{Ctx: ctx})
	must(initError)

	// our "event loop". an event occurs whenever the list of upstreams has been updated
	for {
		select {
		// if we error during watch, just exit
		case err := <-watchErrors:
			must(err)
		// process a new upstream list
		case newUpstreamList := <-upstreamWatch:
			// we received a new list of upstreams from our watch, 
			resync(ctx, newUpstreamList, proxyClient)
		}
	}
}

Finished Code

Great! Here’s what our completed main file should look like:

Click to see the full main file

Run

While it’s possible to package this application in a Docker container and deploy it as a pod inside of Kubernetes, let’s just try running it locally. Make sure you have Gloo Edge installed in your cluster so that Discovery will create some Upstreams for us.


go run example/proxycontroller/proxycontroller.go


kubectl apply -f https://raw.githubusercontent.com/solo-io/gloo/master/example/proxycontroller/install/proxycontroller.yaml

The logs should show:

2019/02/11 11:27:30 wrote proxy object: listeners:<name:"my-amazing-listener" bind_address:"::" bind_port:8080 http_listener:<virtual_hosts:<name:"default-kubernetes-443" domains:"default-kubernetes-443" routes:<matchers:<prefix:"/" > route_action:<single:<upstream:<name:"default-kubernetes-443" namespace:"gloo-system" > > > > > virtual_hosts:<name:"gloo-system-gateway-proxy-8080" domains:"gloo-system-gateway-proxy-8080" routes:<matchers:<prefix:"/" > route_action:<single:<upstream:<name:"gloo-system-gateway-proxy-8080" namespace:"gloo-system" > > > > > virtual_hosts:<name:"gloo-system-gloo-9977" domains:"gloo-system-gloo-9977" routes:<matchers:<prefix:"/" > route_action:<single:<upstream:<name:"gloo-system-gloo-9977" namespace:"gloo-system" > > > > > virtual_hosts:<name:"kube-system-kube-dns-53" domains:"kube-system-kube-dns-53" routes:<matchers:<prefix:"/" > route_action:<single:<upstream:<name:"kube-system-kube-dns-53" namespace:"gloo-system" > > > > > virtual_hosts:<name:"kube-system-tiller-deploy-44134" domains:"kube-system-tiller-deploy-44134" routes:<matchers:<prefix:"/" > route_action:<single:<upstream:<name:"kube-system-tiller-deploy-44134" namespace:"gloo-system" > > > > > > > status:<> metadata:<name:"my-cool-proxy" namespace:"gloo-system" resource_version:"455073" >

Neat! Our proxy got created. We can view it with kubectl:

kubectl get proxy -n gloo-system -o yaml
Click to see the expected proxy

Cool. Let’s leave our controller running and watch it dynamically respond when we add a service to our cluster:

kubectl apply -f https://raw.githubusercontent.com/solo-io/gloo/v1.2.9/example/petstore/petstore.yaml

See the service and pod:

kubectl get pod -n default && kubectl get svc -n default

NAME                      READY     STATUS    RESTARTS   AGE
petstore-6fd84bc9-zdskz   1/1       Running   0          5s
NAME         TYPE        CLUSTER-IP      EXTERNAL-IP   PORT(S)    AGE
kubernetes   ClusterIP   10.96.0.1       <none>        443/TCP    6d
petstore     ClusterIP   10.109.34.250   <none>        8080/TCP   5s

The upstream that was created:

kubectl get upstream -n gloo-system

NAME                              AGE
default-kubernetes-443            2m
default-petstore-8080             46s # <- this one's new
gloo-system-gateway-proxy-8080    2m
gloo-system-gloo-9977             2m
kube-system-kube-dns-53           2m
kube-system-tiller-deploy-44134   2m

And check that our proxy object was updated:

kubectl get proxy -n gloo-system -o yaml

apiVersion: v1
items:
- apiVersion: gloo.solo.io/v1
  kind: Proxy
  metadata:
    creationTimestamp: "2019-12-20T16:29:37Z"
    generation: 12
    name: my-cool-proxy
    namespace: gloo-system
    resourceVersion: "2081"
    selfLink: /apis/gloo.solo.io/v1/namespaces/gloo-system/proxies/my-cool-proxy
    uid: 58e238af-5b1f-4138-bb63-b8edc2405a6c
  spec:
    listeners:
    - bindAddress: '::'
      bindPort: 8080
      httpListener:
        virtualHosts:
        - domains:
          - default-petstore-8080
          name: default-petstore-8080
          routes:
          - matchers:
            - prefix: /
            routeAction:
              single:
                upstream:
                  name: default-petstore-8080
                  namespace: gloo-system
            ... # <- other virtual hosts omitted for brevity
      name: my-amazing-listener
  status: {}
kind: List
metadata:
  resourceVersion: ""
  selfLink: ""

The proxy should have been created with the default-petstore-8080 virtualHost.

Now that we have a proxy called my-cool-proxy, Gloo Edge will be serving xDS configuration that matches this proxy CRD. However, we don’t actually have an Envoy instance deployed that will receive this config. In the next section, we’ll walk through the steps to deploy an Envoy pod wired to receive config from Gloo Edge, identifying itself as my-cool-proxy.

Deploying Envoy to Kubernetes

Gloo Edge comes pre-installed with at least one proxy depending on your setup: the gateway-proxy. This proxy is configured by the gateway proxy controller. It’s not very different from the controller we just wrote!

We’ll need to deploy another proxy that will register to Gloo Edge with it’s role configured to match the name of our proxy CRD, my-cool-proxy. Let’s do it!

If you’d prefer, you can deploy the my-cool-proxy envoy configmap, service, and deployment in one step:

kubectl apply -f https://raw.githubusercontent.com/solo-io/gloo/master/example/proxycontroller/install/my-cool-proxy.yaml

and then skip to testing the proxy.

Creating the ConfigMap

Envoy needs a ConfigMap which points it at Gloo Edge as its configuration server. Run the following command to create the configmap you’ll need:

cat << EOF | kubectl apply -f -
apiVersion: v1
kind: ConfigMap
metadata:
  name: my-cool-envoy-config
  namespace: default
data:
  envoy.yaml: |
    node:
      cluster: "1"
      id: "1"
      metadata:

        # this line is what connects this envoy instance to our Proxy crd
        role: "gloo-system~my-cool-proxy"

    static_resources:
      clusters:
      - name: xds_cluster
        connect_timeout: 5.000s
        load_assignment:
          cluster_name: xds_cluster
          endpoints:
          - lb_endpoints:
            - endpoint:
                address:
                  socket_address:

                    # here's where we provide the hostname of the gloo service
                    address: gloo.gloo-system.svc.cluster.local

                    port_value: 9977
        http2_protocol_options: {}
        type: STRICT_DNS
    dynamic_resources:
      ads_config:
        api_type: GRPC
        grpc_services:
        - envoy_grpc: {cluster_name: xds_cluster}
      cds_config:
        ads: {}
      lds_config:
        ads: {}
    admin:
      access_log_path: /dev/null
      address:
        socket_address:
          address: 127.0.0.1
          port_value: 19000
EOF

Note that this will create the configmap in the default namespace, but you can run it anywhere. Just make sure the proxy deployment and service all go to the same namespace.

Creating the Service and Deployment

We need to create a LoadBalancer service for our proxy so we can connect to it from the outside. Note that if you’re using a Kubernetes Cluster without an external load balancer (e.g. minikube), we’ll be using the service’s NodePort to connect.

Run the following command to create the service:

cat << EOF | kubectl apply -f -
apiVersion: v1
kind: Service
metadata:
  labels:
    gloo: my-cool-proxy
  name: my-cool-proxy
  namespace: default
spec:
  ports:
  - port: 8080 # <- this port should match the port for the HttpListener in our Proxy CRD
    protocol: TCP
    name: http
  selector:
    gloo: my-cool-proxy
  type: LoadBalancer
EOF

Finally we’ll want to create the deployment itself which will launch a pod with Envoy running inside.

cat << EOF | kubectl apply -f -
apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    gloo: my-cool-proxy
  name: my-cool-proxy
  namespace: default
spec:
  replicas: 
  selector:
    matchLabels:
      gloo: my-cool-proxy
  template:
    metadata:
      labels:
        gloo: my-cool-proxy
    spec:
      containers:
      - args: ["--disable-hot-restart"]
        env:
        - name: POD_NAMESPACE
          valueFrom:
            fieldRef:
              fieldPath: metadata.namespace
        - name: POD_NAME
          valueFrom:
            fieldRef:
              fieldPath: metadata.name
        # image: quay.io/solo-io/gloo-envoy-wrapper:1.2.12 # <- use this instead if using open-source Gloo Edge
        image: quay.io/solo-io/gloo-ee-envoy-wrapper:1.2.0 # <- you must use closed-source envoy if using Gloo Edge Enterprise
        imagePullPolicy: Always
        name: my-cool-proxy
        ports:
        - containerPort: 8080 # <- this port should match the port for the HttpListener in our Proxy CRD
          name: http
          protocol: TCP
        volumeMounts:
        - mountPath: /etc/envoy
          name: envoy-config
      volumes:
      - configMap:
          name: my-cool-envoy-config
        name: envoy-config
EOF

If all went well, we should see our pod starting successfully in default (or whichever namespace you picked):

kubectl get pod -n default

NAME                             READY     STATUS    RESTARTS   AGE
my-cool-proxy-7bcb58c87d-h4292   1/1       Running   0          3s
petstore-6fd84bc9-zdskz          1/1       Running   0          48m

Testing the Proxy

If you have glooctl installed, we can grab the HTTP endpoint of the proxy with the following command:

glooctl proxy url -n default --name my-cool-proxy

returns

http://192.168.99.150:30751

Using curl, we can connect to any service in our cluster by using the correct Host header:

curl $(glooctl proxy url -n default --name my-cool-proxy)/api/pets -H "Host: default-petstore-8080"

returns

[{"id":1,"name":"Dog","status":"available"},{"id":2,"name":"Cat","status":"pending"}]

Try any Host header for any upstream name:

kubectl get upstream -n gloo-system

NAME                              AGE
default-kubernetes-443            55m
default-my-cool-proxy-8080        5m
default-petstore-8080             53m
gloo-system-gateway-proxy-8080    55m
gloo-system-gloo-9977             54m
kube-system-kube-dns-53           54m
kube-system-tiller-deploy-44134   54m

Sweet! You’re an official Gloo Edge developer! You’ve just seen how easy it is to extend Gloo Edge to service one of many potential use cases. Take a look at our API Reference Documentation to learn about the wide range of configuration options Proxies expose such as request transformation, SSL termination, serverless computing, and much more.

Appendix - Auto-Generated Routes for Discovered Consul Services

The rest of this guide assumes you’ve been running on minikube, but this setup can be extrapolated to production/more complex setups.

Run consul on your local machine:

consul agent -dev --client=0.0.0.0

Get the Host IP address where consul will be reachable from within minikube pods:

minikube ssh "route -n | grep ^0.0.0.0 | awk '{ print \$2 }'"

Enable consul service discovery in Gloo Edge, replacing address with the value you got before:

kubectl patch settings -n gloo-system default --patch '{"spec": {"consul": {"address": "10.0.2.2:8500", "serviceDiscovery": {}}}}' --type=merge

Get the cluster IP for petstore service and register a consul service to point there:

PETSTORE_IP=$(kubectl get svc -n default petstore -o=jsonpath='{.spec.clusterIP}')
cat > petstore-service.json <<EOF
{
  "ID": "petstore1",
  "Name": "petstore",
  "Address": "${PETSTORE_IP}",
  "Port": 8080
}
EOF

Register the consul service:

curl -v \
    -XPUT \
    --data @petstore-service.json \
    "http://127.0.0.1:8500/v1/agent/service/register"

Confirm the upstream was discovered:

kubectl get us -n gloo-system petstore

returns

NAME       AGE
petstore   8s

Hit the discovered consul service:

curl $(glooctl proxy url -n default --name my-cool-proxy)/api/pets -H "Host: petstore"

returns

[{"id":1,"name":"Dog","status":"available"},{"id":2,"name":"Cat","status":"pending"}]

Nice. You’ve configured Gloo Edge to proactively create routes to discovered consul services!