» Running Consul on Kubernetes

Consul can run directly on Kubernetes, both in server or client mode. For pure-Kubernetes workloads, this enables Consul to also exist purely within Kubernetes. For heterogeneous workloads, Consul agents can join a server running inside or outside of Kubernetes.

This page starts with a large how-to section for various specific tasks. To learn more about the general architecture of Consul on Kubernetes, scroll down to the architecture section. If you would like to get hands-on experience testing Consul on Kubernetes, try the step-by-step beginner tutorial with an accompanying video in the Minikube with Consul guide

» Helm Chart

The recommended way to run Consul on Kubernetes is via the Helm chart. This will install and configure all the necessary components to run Consul. The configuration enables you to run just a server cluster, just a client cluster, or both. Using the Helm chart, you can have a full Consul deployment up and running in seconds.

While the Helm chart exposes dozens of useful configurations and automatically sets up complex resources, it does not automatically operate Consul. You are still responsible for learning how to monitor, backup, upgrade, etc. the Consul cluster.

The Helm chart has no required configuration and will install a Consul cluster with sane defaults out of the box. Prior to going to production, it is highly recommended that you learn about the configuration options.

» How-To

» Installing Consul

To install Consul, clone the consul-helm repository, checkout the latest release, and install Consul. You can run helm install with the --dry-run flag to see the resources it would configure. In a production environment, you should always use the --dry-run flag prior to making any changes to the Consul cluster via Helm.

# Clone the chart repo
$ git clone https://github.com/hashicorp/consul-helm.git
$ cd consul-helm

# Checkout a tagged version
$ git checkout v0.1.0

# Run Helm
$ helm install --name consul ./
...

That's it. The Helm chart does everything to setup a recommended Consul-on-Kubernetes deployment. In a couple minutes, a Consul cluster will be formed and a leader elected and every node will have a running Consul agent.

The defaults will install both server and client agents. To install only one or the other, see the chart configuration values.

» Viewing the Consul UI

The Consul UI is enabled by default when using the Helm chart. For security reasons, it isn't exposed via a Service by default so you must use kubectl port-forward to visit the UI. Once the port is forwarded as shown below, navigate your browser to http://localhost:8500.

$ kubectl port-forward consul-server-0 8500:8500
...

The UI can also be exposed via a Kubernetes Service. To do this, configure the ui.service chart values.

» Joining an Existing Consul Cluster

If you have a Consul cluster already running, you can configure your Kubernetes nodes to join this existing cluster.

global:
  enabled: false

client:
  enabled: true
  join:
    - "provider=my-cloud config=val ..."

The values.yaml file to configure the Helm chart sets the proper configuration to join an existing cluster.

The global.enabled value first disables all chart components by default so that each component is opt-in. This allows us to only setup the client agents. We then opt-in to the client agents by setting client.enabled to true.

Next, client.join is set to an array of valid -retry-join values. In the example above, a fake cloud auto-join value is specified. This should be set to resolve to the proper addresses of your existing Consul cluster.

» Accessing the Consul HTTP API

The Consul HTTP API should be accessed by communicating to the local agent running on the same node. While technically any listening agent (client or server) can respond to the HTTP API, communicating with the local agent has important caching behavior, and allows you to use the simpler /agent endpoints for services and checks.

For Consul installed via the Helm chart, a client agent is installed on each Kubernetes node. This is explained in the architecture section. To access the agent, you may use the downward API.

An example pod specification is shown below. In addition to pods, anything with a pod template can also access the downward API and can therefore also access Consul: StatefulSets, Deployments, Jobs, etc.

apiVersion: v1
kind: Pod
metadata:
  name: consul-example
spec:
  containers:
    - name: example
      image: "consul:latest"
      env:
        - name: HOST_IP
          valueFrom:
            fieldRef:
              fieldPath: status.hostIP
      command:
        - "/bin/sh"
        - "-ec"
        - |
            export CONSUL_HTTP_ADDR="${HOST_IP}:8500"
            consul kv put hello world
  restartPolicy: Never

An example Deployment is also shown below to show how the host IP can be accessed from nested pod specifications:

apiVersion: apps/v1
kind: Deployment
metadata:
  name: consul-example-deployment
spec:
  replicas: 1
  selector:
    matchLabels:
      app: consul-example
  template:
    metadata:
      labels:
        app: consul-example
    spec:
      containers:
        - name: example
          image: "consul:latest"
          env:
            - name: HOST_IP
              valueFrom:
                fieldRef:
                  fieldPath: status.hostIP
          command:
            - "/bin/sh"
            - "-ec"
            - |
                export CONSUL_HTTP_ADDR="${HOST_IP}:8500"
                consul kv put hello world

» Upgrading Consul on Kubernetes

To upgrade Consul on Kubernetes, we follow the same pattern as generally upgrading Consul, except we can use the Helm chart to step through a rolling deploy. It is important to understand how to generally upgrade Consul before reading this section.

Upgrading Consul on Kubernetes will follow the same pattern: each server will be updated one-by-one. After that is successful, the clients will be updated in batches.

» Upgrading Consul Servers

To initiate the upgrade, change the server.image value to the desired Consul version. For illustrative purposes, the example below will use consul:123.456. Also set the server.updatePartition value equal to the number of server replicas:

server:
  image: "consul:123.456"
  replicas: 3
  updatePartition: 3

The updatePartition value controls how many instances of the server cluster are updated. Only instances with an index greater than the updatePartition value are updated (zero-indexed). Therefore, by setting it equal to replicas, none should update yet.

Next, run the upgrade. You should run this with --dry-run first to verify the changes that will be sent to the Kubernetes cluster.

$ helm upgrade consul ./
...

This should cause no changes (although the resource will be updated). If everything is stable, begin by decreasing the updatePartition value by one, and running helm upgrade again. This should cause the first Consul server to be stopped and restarted with the new image.

Wait until the Consul server cluster is healthy again (30s to a few minutes) then decrease updatePartition and upgrade again. Continue until updatePartition is 0. At this point, you may remove the updatePartition configuration. Your server upgrade is complete.

» Upgrading Consul Clients

With the servers upgraded, it is time to upgrade the clients. To upgrade the clients, set the client.image value to the desired Consul version. Then, run helm upgrade. This will upgrade the clients in batches, waiting until the clients come up healthy before continuing.

» Architecture

We recommend running Consul on Kubernetes with the same general architecture as running it anywhere else. There are some benefits Kubernetes can provide that eases operating a Consul cluster and we document those below. The standard production deployment guide is still an important read even if running Consul within Kubernetes.

Each section below will outline the different components of running Consul on Kubernetes and an overview of the resources that are used within the Kubernetes cluster.

» Server Agents

The server agents are run as a StatefulSet, using persistent volume claims to store the server state. This also ensures that the node ID is persisted so that servers can be rescheduled onto new IP addresses without causing issues. The server agents are configured with anti-affinity rules so that they are placed on different nodes. A readiness probe is configured that marks the pod as ready only when it has established a leader.

A Service is registered to represent the servers and expose the various ports. The DNS address of this service is used to join the servers to each other without requiring any other access to the Kubernetes cluster. The service is configured to publish non-ready endpoints so that it can be used for joining during bootstrap and upgrades.

Additionally, a PodDisruptionBudget is configured so the Consul server cluster maintains quorum during voluntary operational events. The maximum unavailable is (n/2)-1 where n is the number of server agents.

» Client Agents

The client agents are run as a DaemonSet. This places one agent (within its own pod) on each Kubernetes node. The clients expose the Consul HTTP API via a static port (default 8500) bound to the host port. This enables all other pods on the node to connect to the node-local agent using the host IP that can be retrieved via the Kubernetes downward API. See accessing the Consul HTTP API for an example.

There is a major limitation to this: there is no way to bind to a local-only host port. Therefore, any other node can connect to the agent. This should be considered for security. For a properly production-secured agent with TLS and ACLs, this is safe.

Some people prefer to run Consul agent per pod architectures, since this makes it easy to register the pod as a service. However, this turns a pod into a "node" in Consul and also causes an explosion of resource usage since every pod needs a Consul agent. We recommend instead running an agent (in a dedicated pod) per node, via the DaemonSet. This maintains the node equivalence in Consul. Service registration should be handled via the catalog syncing feature with Services rather than pods.