10 min • read

Linkerd 2 integration

Linkerd 2 is a zero-config and ultra-lightweight service mesh. Emissary-ingress natively supports Linkerd 2 for service discovery, end-to-end TLS (including mTLS between services), and (with Linkerd 2.8) multicluster operation.

Architecture

Linkerd 2 is designed for simplicity, security, and performance. In the cluster, it runs a control plane in its own namespace and then injects sidecar proxy containers in every Pod that should be meshed.

Emissary-ingress itself also needs to be interwoven or "meshed" with Linkerd 2, and then configured to add special Linkerd headers to requests that tell Linkerd 2 where to forward them. This ie because mTLS between services is automatically handled by the control plane and the proxies. Istio and Consul allow Emissary-ingress to initiate mTLS connections to upstream services by grabbing a certificate from a Kubernetes Secret. However, Linkerd 2 does not work this way, so Emissary-ingress must rely on Linkerd 2 for mTLS connections to upstream services. This means we want Linkerd 2 to inject its sidecar into Emissary-ingress's pods, but not Istio and Consul.

Through that setup, Emissary-ingress terminates external TLS as the gateway and traffic is then immediately wrapped into mTLS by Linkerd 2 again. Thus we have a full end-to-end TLS encryption chain.

Getting started

In this guide, you will use Linkerd 2 Auto-Inject to mesh a service and use Emissary-ingress to dynamically route requests to that service based on Linkerd 2's service discovery data. If you already have Emissary-ingress installed, you will just need to install Linkerd 2 and deploy your service.

Setting up Linkerd 2 requires to install three components. The first is the CLI on your local machine, the second is the actual Linkerd 2 control plane in your Kubernetes Cluster. Finally, you have to inject your services' Pods with Linkerd Sidecars to mesh them.

  1. Install and configure Linkerd 2 instructions. Follow the guide until Step 3. That should give you the CLI on your machine and all required pre-flight checks.

    In a nutshell, these steps boil down to the following:

  2. Now it is time to install Linkerd 2 itself. To do so execute the following command:

    This will install Linkerd 2 in your cluster. For more details on installing Linkerd visit their docs.

    Note that this simple command automatically enables mTLS by default and registers the AutoInject Webhook with your Kubernetes API Server. You now have a production-ready Linkerd 2 setup rolled out into your cluster!

  3. Deploy Emissary-ingress. This howto assumes you have already followed the Emissary-ingress Getting Started guide. If you haven't done that already, you should do that now.

  4. Configure Emissary-ingress to add it to the Linkerd 2 service mesh.

    This will tell Emissary-ingress to add additional headers to each request forwarded to Linkerd 2 with information about where to route this request to. This is a general setting. You can also set add_linkerd_headers per Mapping.

Routing to Linkerd 2 services

You'll now register a demo application with Linkerd 2, and show how Emissary-ingress can route to this application using endpoint data from Linkerd 2.

  1. Enable AutoInjection on the Namespace you are about to deploy to:

    Save the above to a file called namespace.yaml and run kubectl apply -f namespace.yaml. This will enable the namespace to be handled by the AutoInjection Webhook of Linkerd 2. Every time something is deployed to that namespace, the deployment is passed to the AutoInject Controller and injected with the Linkerd 2 proxy sidecar automatically.

  2. Deploy the QOTM demo application. You may have already done this as part of the getting started guide, if so, restart the application using the rollout restart command provided below.

    Save the above to a file called qotm.yaml and deploy it with

    If you already had qotm deployed please restart it with

    Watch via kubectl get pod -w as the Pod is created. Note that it starts with 0/2 containers automatically, as it has been auto-injected by the Linkerd 2 Webhook.

  3. Verify the QOTM pod has been registered with Linkerd 2. You can verify the QOTM pod is registered correctly by accessing the Linkerd 2 Dashboard.

    Your browser should automatically open the correct URL. Otherwise, note the output from the above command and open that in a browser of your choice.

  4. Create a Mapping for the qotm-Linkerd2 service.

    Save the above YAML to a file named qotm-mapping.yaml, and apply it with:

    to apply this configuration to your Kubernetes cluster. Note that in the above config there is nothing special to make it work with Linkerd 2. The general config for Emissary-ingress already adds Linkerd Headers when forwarding requests to the service mesh.

  5. Send a request to the qotm-Linkerd2 API.

Congratulations! You're successfully routing traffic to the QOTM application, the location of which is registered in Linkerd 2. The traffic to Emissary-ingress is not TLS secured, but from Emissary-ingress to the QOTM an automatic mTLS connection is being used.

If you now configure TLS termination in Emissary-ingress, you have an end-to-end secured connection.

Multicluster operation

Linkerd 2.8 can support multicluster operation, where the Linkerd mesh transparently bridges from one cluster to another, allowing seamless access between the two. This works using the Linkerd "service mirror controller" to discover services in the target cluster, and expose (mirror) them in the source cluster. Requests to mirrored services in the source cluster are transparently proxied via Emissary-ingress in the target cluster to the appropriate target service, using Linkerd's automatic mTLS to protect the requests in flight between clusters. By configuring Linkerd to use the existing Emissary-ingress as the ingress gateway between clusters, you eliminate the need to deploy and manage an additional ingress gateway.

Initial multicluster setup

  1. Install Emissary-ingress and the Linkerd multicluster control plane. Make sure you've also linked the clusters.

  2. Inject Emissary-ingress deployment with Linkerd (even if you have AutoInject enabled):

    (It's important to require identity on the gateway port so that automatic mTLS works, but it's also important to let Emissary-ingress handle its own ports. AutoInject can't handle this on its own.)

  3. Configure Emissary-ingress as normal for your application.

At this point, your Emissary-ingress installation should work fine with multicluster Linkerd as a source cluster: you can configure Linkerd to bridge to a target cluster, and all should be well.

Using the cluster as a target cluster

Allowing the Emissary-ingress installation to serve as a target cluster requires explicitly giving permission for Linkerd to mirror services from the cluster, and explicitly telling Linkerd to use Emissary-ingress as the target gateway.

  1. Configure the target cluster Emissary-ingress to allow insecure routing.

    When Emissary-ingress is running in a Linkerd mesh, Linkerd provides transport security, so connections coming in from the Linkerd in the source cluster will always be HTTP when they reach Emissary-ingress. Therefore, the Host CRDs corresponding to services that you'll be accessing from the source cluster must be configured to Route insecure requests. More information on this topic is available in the Host documentation; an example might be

  2. Configure the target cluster Emissary-ingress to support Linkerd health checks.

    Multicluster Linkerd does its own health checks beyond what Kubernetes does, so a Mapping is needed to allow Linkerd's health checks to succeed:

    When configuring Emissary-ingress, Kubernetes is usually configured to run health checks directly against port 8877 -- however, that port is not meant to be exposed outside the cluster. The Mapping permits accessing the health check endpoint without directly exposing the port.

    (The actual prefix in the Mapping is not terribly important, but it needs to match the metadata supplied to the service mirror controller, below.)

  3. Configure the target cluster Emissary-ingress for the service mirror controller.

    This requires changes to the Emissary-ingress's deployment and service. For all of these commands, you will need to make sure your Kubernetes context is set to talk to the target cluster.

    In the deployment, you need the config.linkerd.io/enable-gateway annotation:

    In the service, you need to provide appropriate named port definitions:

    • mc-gateway needs to be defined as port 4143
    • mc-probe needs to be defined as port 80, targetPort 8080 (or wherever Emissary-ingress is listening)

    Finally, the service also needs its own set of annotations:

    (Here, the value of mirror.linkerd.io/probe-path must match the prefix using for the probe Mapping above.)

  4. Configure individual exported services. Adding the following annotations to a service will tell the service to use Emissary-ingress as the gateway:

    This annotation will tell Linkerd that the given service can be reached via the Emissary-ingress in the emissary namespace.

  5. Verify that all is well from the source cluster.

    For all of these commands, you'll need to set your Kubernetes context for the source cluster.

    First, check to make that the clusters are correctly linked:

    Next, make sure that the Emissary-ingress gateway shows up when listing active gateways:

    At this point, all should be well!

More information

For more about Emissary-ingress's integration with Linkerd 2, read the service discovery configuration documentation. For further reading about Linkerd 2 multi-cluster, see the install documentation and introduction.