Deployment Process

The deployment process on Scalingo is composed of several steps. This article aims at describing each of those to offer a comprehensive understanding to the reader.

Deployment Trigger

The first step to trigger a deployment is to send the application itself to the platform (either the code or a build version of it). Multiple methods are available to achieve this step:

Special deployment methods are available for JVM-based applications:

Building the application

Once the deployment has been triggered, the next phase is named build. During this step, the deployment system detects the technologies used by the application and installs everything required in order to make it runnable.

For instance, in the case of a Ruby on Rails application, the correct version of Ruby is installed, then all the gems are downloaded using bundle and the assets are precompiled.

This process to detect the technology and to install all its dependencies is based on open-source tools named buildpack. At the beginning of the build, the officially supported buildpacks are automatically used to detect which one can be used to build the application, they usually cover most use cases.

However it may happen your application technology is not detected, or your application is using multiple stacks (ie. Python + NodeJS), both these cases are also covered.

As mentionned before buildpacks are open-source and other hosting providers are also using them to deploy applications, thus the developer community has developed a wide range of buildpacks for almost any kind of technology. The platform has been designed to be extended by community buildpacks, learn how to do it here.

If a project is based on multiple language runtimes, buildpacks can be combined to install dependencies of multiple stacks. A custom buildpack has been created to achieve this goal: the multi buildpacks.

Rolling out the new version of the application

Once your code is built, the deployment arrives at its last step, aiming at replacing the previous version of the application by the one which just got built. This step ensures the platform is able to achieve zero-downtime/rolling deployments of application. The process is the following:

  1. Start containers with the new version of the application
  2. Wait them to be ready: TCP SYN are sent to web and tcp containers, to check if the process inside them has correctly started and is listening to the port defined by the environment variable PORT. More details about container start operations is available here.
  3. If a postdeploy hook is defined, it is executed in a one-off container of the new version of the application. In the case the hook fails, the deployment would stop and the previous version would keep running.
  4. Network traffic is routed to the containers hosting the new version of the application.

Once these 4 steps have ended successfuly, the deployment is considered as successful, and the order to shutdown is sent to old containers. They now have 30 seconds to cleanup cleanly.

schedule 09 Feb 2018