By: Sandeep Dinesh - July 29, 2015
Manually deploying Containers is all fine and dandy, but is rather fragile and clumsy. What happens if the app crashes? How can the app be updated? Rolled back?
Thankfully, there is a system we can use to manage our containers in a cluster environment called Kubernetes. Even better, Google has a managed version of Kubernetes called Google Container Engine so you can get up and running in minutes.
Before we jump in and start kube’ing it up, it’s important to understand some of the fundamentals of Kubernetes.
In my previous post, I used off-the-shelf containers to keep things simple.
I had a stock MongoDB container and a stock Node.js container. The Mongo container ran fine without any modification. However, I had to manually enter the Node container to pull and run the code. Obviously this isn’t ideal in Kubernetes land, as you aren’t supposed to log into your servers!
Instead, you have to build a custom container that has the code already inside it and runs automatically.
To do this, you need to use more Docker. Make sure you have the latest version installed for the rest of this tutorial.
Getting the code:
Before starting, let’s get some code to run. You can follow along on your personal machine or a Linux VM in the cloud. I recommend using Linux or a Linux VM; running Docker on Mac and Windows is outside the scope of this tutorial.
$ git clone https://github.com/ijason/NodeJS-Sample-App.git app $ mv app/EmployeeDB/* app/ $ sed -i -- 's/localhost/mongo/g' ./app/app.js
This is the same sample app we ran before. The second line just moves everything from the
EmployeeDB subfolder up into the app folder so it’s easier to access. The third line, once again, replaces the hardcoded
localhost with the
Building the Docker image:
First, you need a
Dockerfile. This is basically the list of instructions Docker uses to build a container image.
Here is the
Dockerfile for the web server:
FROM node:4.4 RUN mkdir -p /usr/src/app WORKDIR /usr/src/app COPY ./app/ ./ RUN npm install CMD ["node", "app.js"]
Dockerfile is pretty self-explanatory, and this one is dead simple.
First, it uses the official Node.js LTS image as the base image.
Then, it creates a folder to store the code,
cds into that directory, copies the code in, and installs the dependencies with npm.
Finally, it specifies the command Docker should run when the container starts, which is to start the app.
Right now, the directory should look like this:
$ ls Dockerfile app
$ docker build -t myapp .
This will build a new Docker image for your app. This might take a few minutes as it is downloading and building everything.
After that is done, test it out:
$ docker run myapp
At this point, you should have a server running on
http://localhost:3000 (or wherever Docker tells you). The website will error out as there is no database running, but we know it works!
Now you have a custom Docker image, you have to actually access it from the cloud.
As we are going to be using the image with Google Container Engine, the best place to push the image is the Google Container Registry. The Container Registry is built on top of Google Cloud Storage, so you get the advantage of scalable storage and very fast access from Container Engine.
First, make sure you have the latest version of the Google Cloud SDK installed.
$ curl https://sdk.cloud.google.com | bash
Then, make sure you log in and update.
$ gcloud auth login $ gcloud components update
You’re ready to push your container live, but you’ll need a destination. Create a Project in the Google Cloud Platform Console, and leave it blank. Use the Project ID below, and push your project live.
$ docker tag myapp gcr.io/<YOUR-PROJECT-ID>/myapp $ gcloud docker push gcr.io/<YOUR-PROJECT-ID>/myapp
After some time, it will finish. You can check the console to see the container has been pushed up.
So now you have the custom container, let’s create a cluster to run it.
Currently, a cluster can be as small as one machine to as big as 100 machines. You can pick any machine type you want, so you can have a cluster of a single
f1-micro instance, 100
n1-standard-32 instances (3,200 cores!), and anything in between.
For this tutorial I’m going to use the following:
us-central-1f(Use a zone close to you)
There are two ways to create this cluster. Take your pick.
$ gcloud beta container \ --project "<YOUR-PROJECT-ID>" \ clusters create "mean-cluster" \ --zone "us-central1-f" \ --machine-type "n1-standard-1" \ --num-nodes "2" \ --network "default"
After a few minutes, you should see this in the console.
Three things need to be created:
To create the disk, run this:
$ gcloud compute disks create \ --project "<YOUR-PROJECT-ID>" \ --zone "us-central1-f" \ --size 200GB \ mongo-disk
Pick the same zone as your cluster and an appropriate disk size for your application.
Now, we need to create a Deployment that will run the database. I’m using a Deployment and not a Pod, because if a standalone Pod dies, it won’t restart automatically.
apiVersion: extensions/v1beta1 kind: Deployment metadata: name: mongo-deployment spec: replicas: 1 template: metadata: labels: name: mongo spec: containers: - image: mongo name: mongo ports: - name: mongo containerPort: 27017 hostPort: 27017 volumeMounts: - name: mongo-persistent-storage mountPath: /data/db volumes: - name: mongo-persistent-storage gcePersistentDisk: pdName: mongo-disk fsType: ext4
We call the deployment
mongo-deployment, specify one replica, and open the appropriate ports. The image is
mongo, which is the off the shelf MongoDB image.
volumes section creates the volume for Kubernetes to use. There is a Google Container Engine-specific
gcePersistentDisk section that maps the disk we made into a Kubernetes volume, and we mount the volume into the
/data/db directory (as described in the MongoDB Docker documentation)
Now we have the Deployment, let’s create the Service:
apiVersion: v1 kind: Service metadata: labels: name: mongo name: mongo spec: ports: - port: 27017 targetPort: 27017 selector: name: mongo
Again, pretty simple stuff. We “select” the mongo Deployment to be served, open up the ports, and call the service
This is just like the “link” command line option we used with Docker in my previous post. Instead of connecting to
localhost, we connect to
mongo, and Kubernetes redirects traffic to the mongo service!
At this point, the local directory looks like this:
$ ls Dockerfile app db-deployment.yml db-service.yml
First, let’s “log in” to the cluster
$ gcloud container clusters get-credentials mean-cluster
Now create the Deployment.
$ kubectl create -f db-deployment.yml
And the Service.
$ kubectl create -f db-service.yml
kubectl is the Kubernetes command line tool (automatically installed with the Google Cloud SDK). We are just creating the resources specified in the files.
At this point, the database is spinning up! You can check progress with the following command:
$ kubectl get pods
Once you see the mongo pod in running status, we are good to go!
$ kubectl get pods NAME READY REASON RESTARTS AGE mongo-deployment-xxxx 1/1 Running 0 3m
Now the database is running, let’s start the web server.
We need two things:
Let’s look at the Deployment configuration:
apiVersion: extensions/v1beta1 kind: Deployment metadata: name: web-deployment spec: replicas: 2 template: metadata: labels: name: web spec: containers: - image: gcr.io/<YOUR-PROJECT-ID>/myapp name: web ports: - name: http-server containerPort: 3000
Here, we create a deployment called
web-deployment, and we tell it to create two replicas. Replicas of what you ask? You may notice the
template section looks just like a Pod configuration, and that’s because it is. We are creating a Pod with our custom Node.js container and exposing port 3000.
Now for the Service
apiVersion: v1 kind: Service metadata: name: web labels: name: web spec: type: LoadBalancer ports: - port: 80 targetPort: 3000 protocol: TCP selector: name: web
Notice two things here:
At this point, the local directory looks like this
$ ls Dockerfile app db-deployment.yml db-service.yml web-service.yml web-deployment.yml
Create the Deployment.
$ kubectl create -f web-deployment.yml
And the Service.
$ kubectl create -f web-service.yml
And check the status.
$ kubectl get pods
Once you see the web pods in running status, we are good to go!
$ kubectl get pods NAME READY REASON RESTARTS AGE mongo-deployment-xxxx 1/1 Running 0 4m web-deployment-xxxx 1/1 Running 0 1m web-deployment-xxxx 1/1 Running 0 1m
At this point, everything is up and running. The architecture looks something like this:
By default, port 80 should be open on the load balancer. In order to find the IP address of our app, run this command:
$ gcloud compute forwarding-rules list NAME REGION IP_ADDRESS IP_PROTOCOL TARGET abcdef us-central1 104.197.XXX.XXX TCP us-xxxx
If you go to the IP address listed, you should see the app up and running!
And the Database works!
By using Container Engine and Kubernetes, we have a very robust, container based MEAN stack running in production.
In another post, I cover how to setup a MongoDB replica set. This is very important for running in production.
Hopefully I can do some more posts about advanced Kubernetes topics such as changing the cluster size and number of Node.js web server replicas, using different environments (dev, staging, prod) on the same cluster, and doing rolling updates.