Docker

Docker can be used to run containers across different managers. Each manager runs their own docker instances, but the cluster will co-ordinate and schedule instances appropriately.

Manager Options

• slots

The available docker scheduling slots on this host.

Endpoint Options

• slots

The number of scheduled slots required.

• image

The docker image name.

• command

The docker command to run in the given image.

• user

The user to use for running the command.

• environment

The environment for the command.

• mem_limit

The memory limit for the container.

• dns

The DNS server for the container.

• hostname

The hostname of the container.

Example

Setting Up Docker

In order to use Reactor with Docker, you’ll first need to install Docker. This is straight-forward if you follow the instructions here.

You’ll also need to ensure that docker-py is installed. You can generally it this as follows.

sudo pip install https://github.com/dotcloud/docker-py/archive/master.zip

Once all the necessary components are installed, restart reactor.

/etc/init.d/reactor restart

Creating an Image

For this example, we assume you’ve setup the python example as specified in the Docker tutorial here.

You can specify any image to use with Reactor, in the same way you would specify an image on the command line when running Docker directly.

Create an Endpoint

To keep things simple, unmanage the default HTTP endpoints if they still exist.

reactor unmanage apihttp
reactor unmanage apihttps

Then, create a simple configuration for our docker application. Note that your image parameter will be equal to the $BUILD_IMG in the above tutorial.

[endpoint]
url=http://
cloud=docker
loadbalancer=nginx
port=5000

[scaling]
min_instances=1
max_instances=5

[cloud:docker]
image=eb2b727f547d
command=/usr/local/bin/runapp

Next, create an endpoint for our docker application.

reactor manage pyex < docker.conf
reactor start pyex

Finally, check that you can access your auto-scaled, distributed docker application.

curl http://127.0.0.1

Using Auto-scaling

As an example, we’ll set the above endpoint to maintain an average of one connection per second per docker instance.

reactor set pyex scaling rules '0 < rate < 1'

Check the current metrics.

reactor get-metrics pyex

You should see:

{u'active': 0.0}

Now, hit the endpoint a lot.

while true; do curl http://localhost; done

After a while, stop and you check the metrics.

reactor get-metrics pyex

You should see something like:

{u'active': 3.6, u'rate': 33.01126731369651, u'bytes': 396.1352077643581, u'response': 0.002031185031185033}

Shortly, Reactor will start more instances. And after the decomission period, they will be killed.