Meteor on Kubernetes

This example shows you how to package and run a Meteor app on Kubernetes.

Get started on Google Compute Engine

Meteor uses MongoDB, and we will use the GCEPersistentDisk type of volume for persistent storage. Therefore, this example is only applicable to Google Compute Engine. Take a look at the volumes documentation for other options.

First, if you have not already done so:

1. Create a Google Cloud Platform project.
2. Enable billing.
3. Install the gcloud SDK.

Authenticate with gcloud and set the gcloud default project name to point to the project you want to use for your Kubernetes cluster:

gcloud auth login
gcloud config set project <project-name>

Next, start up a Kubernetes cluster:

wget -q -O - https://get.k8s.io | bash

Please see the Google Compute Engine getting started guide for full details and other options for starting a cluster.

Build a container for your Meteor app

To be able to run your Meteor app on Kubernetes you need to build a Docker container for it first. To do that you need to install Docker Once you have that you need to add 2 files to your existing Meteor project Dockerfile and .dockerignore.

Dockerfile should contain the below lines. You should replace the ROOT_URL with the actual hostname of your app.

FROM chees/meteor-kubernetes
ENV ROOT_URL http://myawesomeapp.com

The .dockerignore file should contain the below lines. This tells Docker to ignore the files on those directories when it's building your container.

.meteor/local
packages/*/.build*

You can see an example meteor project already set up at: meteor-gke-example. Feel free to use this app for this example.

Note: The next step will not work if you have added mobile platforms to your meteor project. Check with meteor list-platforms

Now you can build your container by running this in your Meteor project directory:

docker build -t my-meteor .

Pushing to a registry

For the Docker Hub, tag your app image with your username and push to the Hub with the below commands. Replace <username> with your Hub username.

docker tag my-meteor <username>/my-meteor
docker push <username>/my-meteor

For Google Container Registry, tag your app image with your project ID, and push to GCR. Replace <project> with your project ID.

docker tag my-meteor gcr.io/<project>/my-meteor
gcloud docker -- push gcr.io/<project>/my-meteor

Running

Now that you have containerized your Meteor app it's time to set up your cluster. Edit meteor-controller.json and make sure the image: points to the container you just pushed to the Docker Hub or GCR.

We will need to provide MongoDB a persistent Kubernetes volume to store its data. See the volumes documentation for options. We're going to use Google Compute Engine persistent disks. Create the MongoDB disk by running:

gcloud compute disks create --size=200GB mongo-disk

Now you can start Mongo using that disk:

kubectl create -f examples/meteor/mongo-pod.json
kubectl create -f examples/meteor/mongo-service.json

Wait until Mongo is started completely and then start up your Meteor app:

kubectl create -f examples/meteor/meteor-service.json
kubectl create -f examples/meteor/meteor-controller.json

Note that meteor-service.json creates a load balancer, so your app should be available through the IP of that load balancer once the Meteor pods are started. We also created the service before creating the rc to aid the scheduler in placing pods, as the scheduler ranks pod placement according to service anti-affinity (among other things). You can find the IP of your load balancer by running:

kubectl get service meteor --template="{{range .status.loadBalancer.ingress}} {{.ip}} {{end}}"

You will have to open up port 80 if it's not open yet in your environment. On Google Compute Engine, you may run the below command.

gcloud compute firewall-rules create meteor-80 --allow=tcp:80 --target-tags kubernetes-node

What is going on?

Firstly, the FROM chees/meteor-kubernetes line in your Dockerfile specifies the base image for your Meteor app. The code for that image is located in the dockerbase/ subdirectory. Open up the Dockerfile to get an insight of what happens during the docker build step. The image is based on the Node.js official image. It then installs Meteor and copies in your apps' code. The last line specifies what happens when your app container is run.

ENTRYPOINT MONGO_URL=mongodb://$MONGO_SERVICE_HOST:$MONGO_SERVICE_PORT /usr/local/bin/node main.js

Here we can see the MongoDB host and port information being passed into the Meteor app. The MONGO_SERVICE... environment variables are set by Kubernetes, and point to the service named mongo specified in mongo-service.json. See the environment documentation for more details.

As you may know, Meteor uses long lasting connections, and requires sticky sessions. With Kubernetes you can scale out your app easily with session affinity. The meteor-service.json file contains "sessionAffinity": "ClientIP", which provides this for us. See the service documentation for more information.

As mentioned above, the mongo container uses a volume which is mapped to a persistent disk by Kubernetes. In mongo-pod.json the container section specifies the volume:

{
        "volumeMounts": [
          {
            "name": "mongo-disk",
            "mountPath": "/data/db"
          }

The name mongo-disk refers to the volume specified outside the container section:

{
    "volumes": [
      {
        "name": "mongo-disk",
        "gcePersistentDisk": {
          "pdName": "mongo-disk",
          "fsType": "ext4"
        }
      }
    ],