We removed our free Sandbox April 25th.
You can read more on our blog.

How it Works


Now that you have deployed a simple app, and a more involved app with a database, we can review the different steps happening behind a "dotcloud push". That should help you gain a deeper understanding of how the dotCloud platform works, and what’s behind our “zero-downtime pushes” feature.

The Code Store

We started by adding a dotCloud Build File to our code and using the client to push the code to dotCloud. When you use the client to push code, the client will choose the best upload method for your app. For example, if you have a Git or Mercurial repository, the upload will be like a repository push. If you do not use a supported Version Control System, the client will upload your code directly comparing your local code with any previously uploaded code so that only changes are uploaded.


You can learn more about the way the CLI chooses the best upload method and how to override it in the corresponding guide.

Our internal name for the code store is just “uploader”.

The Builder

Once new code has been uploaded, we start looking for your dotcloud.yml Build File, and we deploy the stack of services it describes on our dedicated build cluster. Each service in your stack is built accordingly to a specific set of rules (e.g: a Python service will run "pip" to install dependencies whereas a NodeJS service will run "npm"). In addition to these predefined rules, you can setup hooks to be executed before or after the build. Finally your application is packaged and stored for the deployment phase. You can see the builder in action in the "dotcloud push" output:

---> Building the application...
[www] Build started for revision rsync-1339191773365 (clean build)
[www] I am snapshotsworker_02/bob-2, and I will be your builder today.
[www] Build completed successfully. Compiled image size is 427KB
---> Application build is done

The Deployer

If your application built successfully, then the platform deploys your stack on the Sandbox, Live or Enterprise cluster depending on its flavor.

It starts by initializing a full new stack of service, while the current one (if it’s not your first push on this application) are still running and serving traffic. When the new stack is initialized, the platform retrieves the application package from the builder and install it.

Finally the postinstall hook is run on each service and the new stack is launched. Once this new version of the application is running and ready to accept requests, we seamlessly switch the traffic to it (unless you use a data directory, in that case a short downtime will happen while move it to the new service):

---> Initializing new services... (This may take a few minutes)
---> Using default scaling for service www (1 instance(s)).
[www.0] Initializing...
[www.0] Service initialized
---> All services have been initialized. Deploying code...
[www.0] Deploying build revision rsync-1339191773365...
[www.0] Running postinstall script...
[www.0] Launching...
[www.0] Waiting for the instance to become responsive...
[www.0] Re-routing traffic to the new build...
[www.0] Successfully deployed build revision rsync-1339191773365
---> Deploy finished
---> Application fully deployed

The Stack Runtime

During its lifetime, your application is continually monitored, and services automatically restarted when error conditions occur.

Each service is independent of the others; and, more importantly, services never interact directly with the core platform, except when you deploy (push) new code. That means that your services won’t be impacted when we have to perform maintenance operations on the dotCloud API. Likewise, if you experience slow response times or errors with the dotCloud CLI or website, your websites will not be affected since they are decoupled from those components.

You should now have a much better idea of how to dotCloud works. At this point, you may want to: