In the article Load Balancing with Docker Swarm, we scaled a service by deploying multiple instance of the same docker image across the hosts in a Docker Swarm and distibuted the traffic among these instances using a load balancer. However, the scaling is manually done using
In this article, we are going to automate the scaling procedure using Docker Remote API. We will be creating a
Replicator docker image that listens to requests with container ID as the parameter and can create and deploy new docker images similar to the one with the given container ID.
Docker Remote API
Docker Remote API allows us to remotely access the Docker Engine and do all the actions that we could do locally. You can see the API reference here. For our purposes, we need the following endpoints:
GET /containers/<id>/json: To inspect the container.
POST /containers/create: To create a new container.
POST /containers/<id>/start: To deploy the new container.
To send requests to the Docker Remote API, we need to verify the client using the certificate (cert.pem) and the private key (key.pem) files. To verify the client, we also need the certificate authority (ca.pem) file. When using Docker Machine, the environment variable
DOCKER_CERT_PATH contains the path of the folder containing these files.
Replicator docker image will listen for requests which contains a container ID. This ID is used to inspect the container and create one similar to it. Then the newly created container is deployed. Now we can scale by sending a request to this Replicator image with the container ID. This can be done in two ways:
- A running container can request the Replicator with its own container ID, if it cannot handle the traffic.
- The container can send metrics to some monitoring service along with its container ID. Then the monitoring service can send the request to the Replicator with the container ID as needed.
The Replicator image will be based on the Alpine Linux and it will contain a Node.js server along with NGINX. Node.js server will be used to listen for the request and communicate with Docker Remote API. NGINX is used as a reverse proxy to the Node.js server and it also handle the client verification with the certificates. The docker image for Replicator is hanzel/replicator and its code can be found here.
We need the certificate and key file to access the Remote API. So, we use these same files to authenticate the clients of our Docker Image. We need to have the file
ca.pem in the folder
/ssl in our Docker Image. We will discuss how to get the files there later when we are testing this image. We also need to set the
DOCKER_HOST environment variable with IP and port to access the Remote API. The Docker Machine save this in the format
tcp://<ip>:<port> inside the environment variable
DOCKER_HOST. We just have to pass this value to the docker image. In the next three sections, we will build the Replicator image.
server.js and uses koa as the web server and
co-request to send requests to Docker Remote API. So the
package.json file will look like this.
server.js files starts with setting values to the variables.
We assiged the
koa module to
fs module to
co-request module to
request. The file
/ssl/cert.pem is read and saved to
cert. Similarly, value of
ca is set. We have the IP and port to access the Remote API in the environment variable
DOCKER_HOST in the format
tcp://<ip>:<port>. We strip out the
tcp:// part of it and save it to
host variable. So, if the value of the environment variable
tcp://192.168.99.100:2376, the value of
host will be
Now we have to write the function that handles all the requests. We accept GET request with container ID as the URL parameter. So the request
GET /9d65f58cca99 will be accepted and the container ID is
9d65f58cca99. The function looks like the following.
First of all, we check if there is any URL parameter. If there is no parameter, we send a
400 BAD REQUEST response. Or else, we take the container ID from the URL, save it in
container variable and continue. We send a
GET /containers/<id>/json request to the Docker Remote API to get the configuration of the given image using the
ca variables for authentication. The URL for this is
If the response code for this request is not 200, we send a
404 NOT FOUND response. This happens when the container ID is invalid. If the response code is 200, we continue. Now, we have to send a
POST /containers/create request. The body of this request must contain the configuration of the Docker Image that we wish to create. As we are going to duplicate the image, whose ID was given, this can be obtained from the reponse of the previous request. The body of this request is composed of the
Config object of the previous response.
Now we need to empty the value of
Hostname parameter in the body. This is done so that the newly created image will have a unique hostname which is the substring of its ID. Now we have to set the
HostConfig parameter in the request body. This is be obtained from the
HostConfig object in the previous reponse. Now we send this
POST /containers/create request and the new container is created. The response from this request will contain the ID of the new container in the
Now we send the
POST /containers/<id>/start request to start this newly created container. The ID for this reqeust can be obtained from the previous response, so the URL will be
'https://'+host+'/containers/'+init.body.Id+'/start'. This will start the new container and we can now send the
200 OK response with
done as the body.
Finally, we need to make this app listen to the port
3000 using the command,
app.listen(3000). So the entire
server.js file looks like this.
We are using NGINX as the reverse proxy to our Node.js Application. NGINX is also used for SSL termination and client authentication. The encrypted requests will be received by the NGINX server, the SSL will be terminated and the plain-text requests will be forwarded to our Node.js server running at port
3000. The configuration file
nginx.conf look like this.
NGINX listens to port 443 with SSL turned on. As, port 443 is the default port for SSL, we can now access this service with
https without any ports. We authenticate the clients using the files
ca.pem in the
/ssl folder. We use the
proxy_pass parameter to pass these requests to the Node.js server listening at port
We need to have both Node.js and NGINX to be running inside the docker image. So we will write a script
start.sh which acts as our entry point.
Our Docker Image will be based on Alpine Linux with Node.js server and NGINX running in it. We need to put the four files
start.sh into a folder named
files. Create the
Dockerfile in the folder containing this
files folder and it will contain the following.
The base image is
mhart/alpine-node:6, which is based on Alpine Linux and contains Node.js v6. We install nginx using Alpine Pacakge Manager,
apk. We move the files
server.js to the folder
/code in the image and it will act as the working directory. We run
npm install in this location to install all Node.js dependencies from
Next, we move the file
/etc/nginx folder and
/bin folder. We make the
start.sh executable with
chmod command. The port
443 will be exposed and accessible from outside. This port is listened by NGINX. Finally, we set the
start.sh file as the entry point to this image. You can build this image with the command
docker build -t replicator ..
To send requests to the replicator from inside a running container, the replicator must run in the same network. If the container name of the replicator image is
replicator, we can send request to
https://replicator. We also need to provide the container ID as the URL parameter. This value can be accessed from the environment variable
For authentication of this request, we need to provide the certificate file (cert.pem) and its private key (key.pem). If
curl is installed in the image and the authentication files are in
/ssl folder, we can make this request using the following command.
To test the working of this replicator image, I have made a docker image. This docker image contains a Node.js server that server an HTML page with a button. If the button is clicked, the above
curl command is executed. I have built it into a docker image, hanzel/node-replicate and its code can be found here. With this image, we can see that everytime we click the button a new instance of this service will be spun up. We will be deploying all this to a Docker Swarm now.
Creating the Swarm
We will be using Docker Machine to create and manage remote hosts as a swarm. With Docker Machine, you can create hosts on your local machine or your cloud provider. Check this link to see the drivers supported by Docker Machine.
You need to have the following installed in you local computer:
Docker: version >= 1.10, to support Docker Compose File version 2 and Multi-Host networking.
Docker Machine: version >= 0.6
Docker Compose: version >= 1.6, to support Docker Compose file version 2
The first thing we need to do is to create the Docker Swarm using Docker Machine and set it up. I have explained how to do this in the article, Load Balancing with Docker Swarm. Follow the steps from
Initial Setup to
The Swarm of that article to create and setup the Swarm.
Once the swarm is setup, you can see the hosts with
docker-machine ls command. The output of this command must look something like this.
For the replicator image to access the Docker Remote API, it needs the certificate file (cert.pem), private key (key.pem) and certificate authority file (ca.pem). As we are using Docker Machine, the value in the environment variable
DOCKER_CERT_PATH is the path of the folder containing these files. We are also going to use these same files to authenticate requests coming to the replicator. So we need to have these three files in the
Before moving the files to the Docker images, we need to get the files to all the remote hosts that we create with Docker Machine. We create a new folder name
ssl and copy the required three file to this folder.
We now use
docker-machine scp command to copy the contents of
ssl folder to
/home/ssl folder in the remote hosts
If you have more remote hosts, just repeat these commands by changing the host name. Once these files are in the remote hosts, you can create volumes that point from
/home/ssl in the host to
/ssl in the container, to bring these files into the containers.
Running with Docker Compose
We will test the working of the replicator using Docker Compose. The configuration file
docker-compose.yml looks like the following.
We have an overlay network
web that contains all the services. The first service is the
replicator and uses
hanzel/replicator image. We give the name
replicator for this container and export the port
443. This make the replicator accessible at
https://replicator. We set the
constraint:node environment variable to
slave so that this container always runs in the
slave remote host. As discussed above, we need to set the
DOCKER_HOST environment variable to the URL to access the Remote API. This is set in the
DOCKER_HOST environment variable set by the
docker-machine. We also need to create a Docker Volume from
/home/ssl in the host to
/ssl in the container to share the certificates and keys.
The second service is a load balancer which can distibuted the traffic to different instances of the same image. To know more about this service, read my article Load Balancing with Docker Swarm. The third service is named
web and contains the image
hanzel/node-replicate that was created to test the replicator. It listens to port
3000, so it is exposed. As this service make requests to the replicator, it need the same certificates and keys for authentication. This is done by creating a docker volume similar to
Make sure that your docker client is connected to the swarm with
eval $(docker-machine env -swarm master) command. Now open up the terminal in the folder containing
docker-compose.yml and start the services using the following command.
This will start these services. You can see the running containers with the command
docker-compose ps. The output of the command must look like this.
We can see the application from the url given by the command,
docker-compose port lb 80. You will get some IP address like
184.108.40.206:80. Go to this url and we can see the running app. To test the
replicator, click on the
Replicate button. This sends a request to replicator with the current container’s ID and the replicator will create a new service similar to this and deploy it. Once all that is done, the text
done appears in the webpage.
Check again the running services with
docker-compose ps command and we can see two instances of our
node-replicate images running.
Here, the container named
reverent_dubinsky in the new one replicated. Everytime we press the button, a new instance will be deployed. You can also send the request to replicator externally. You need to have the certificate and key file in the
ssl folder. Run the command
docker ps to get the list of running container. Pick the required container and note its ID. Now run the following command to replicate this container.
Run this command or press the button a few times and the output of
docker-compose ps will look something like this.
You can still use the
docker-compose scale command to manually scale the services. Also, you can see that the load balancer is also updated with the new instances. You can see the load balancer configuration with the command
docker exec -t lb cat /etc/nginx/conf.d/default.conf. Its output looks something like this.
In this article, we can set up an auto-scaling system with Docker using Docker Remote API. This can be used to scale the service as the traffic increases. We have made a docker image to that replicates services and tested it on an app deployed with Docker Swarm.
Once you are done, the services can be stopped and the hosts removed with the following commands.