Introducing a new generation of AWS Elastic Beanstalk platforms

In my last post I discussed AWS Elastic Beanstalk’s new public roadmap on GitHub. Today I want to talk about our new generation of Elastic Beanstalk platforms built on top of Amazon Linux 2 (AL2).

Late last year we launched a public beta of a new Elastic Beanstalk platform for Amazon Corretto — Amazon’s no-cost, production-ready distribution of the Open Java Development Kit (OpenJDK). This is also our first platform based on AL2. This year we have launched two more beta AL2 platforms: Docker and Python. More beta platforms are arriving soon, followed by generally available platform releases.

A sample application using the new Python 3.7 beta platform

I want to dive a little deeper on what we are doing with these platforms. Elastic Beanstalk was publicly launched in 2011, and announced in a blog post by Jeff Barr. Back then there were few enough AWS services that they were all listed as tabs along the top of the AWS Management Console. At launch, we supported only Apache Tomcat applications. Over time, we added support for many other runtimes and began using the term “platform” to describe our offerings. Today we support a wide variety of platforms for popular web application frameworks. For example, Ruby on Rails, PHP, and Node.js, as well as generic Docker-based platforms. In the years since we launched each platform, the underlying communities have continued to evolve. Elastic Beanstalk is an opinionated service, especially when it comes to our platforms. As the service evolves, the opinions baked into our platforms must evolve as well.

With our AL2 platforms, we are refreshing each platform based on feedback we’ve gotten from customers. For example, with Java we heard concerns from many customers about long-term support and licensing of OpenJDK. That’s why in AL2 we are using Amazon’s own Corretto distribution, which includes committed long-term support. It also has performance and scalability improvements learned from Amazon’s years of experience running Java across thousands of production services — such as the Elastic Beanstalk service itself. For more details, see this section of our Java platform documentation.

Our Python AL2 platform has also been modernized. Previously we only supported serving applications through Apache and mod_wsgi. Now we are using NGINX as a reverse proxy in front of Gunicorn, with the flexibility to use another Web Server Gateway Interface (WSGI) server if you prefer. We also took this opportunity to add support for Pipenv and Pipfile, more modern and powerful Python dependency management tools. Learn more in our Python platform documentation.

The Docker AL2 platform is rewritten internally, but provides largely the same customer experience. It does offer improved I/O performance by using the OverlayFS storage driver. This is a change from the previous Docker platform, which used the older and slower Device Mapper storage driver and required an extra Amazon EBS volume.

We are hard at work on another set of beta platforms including PHP, Ruby, and Node.js, which are expected to launch soon. Each of these have been modernized and improved. For a full list of differences between our existing platforms and their Amazon Linux 2 equivalents, check out our documentation. In the next section I want to take a closer look at one new feature that applies to all of the new platforms: platform hooks.

Platform hooks

With our AL2 platforms, we are offering a simplified model for on-instance customization. We’ve long supported configuration files called ebextensions that allow customization of environment options, resources, and on-instance behavior. These have enabled customers to extend their environments in ways we never dreamed of. But we’ve also heard customer feedback about the difficulty of writing complex shell scripts embedded within YAML or JSON. And as they are, ebextensions don’t provide any straightforward mechanism to execute custom code after an application deployment is completed. Customers have pointed out many use cases where they want to do this – for example to enable third party monitoring tools.

With our new generation of Linux platforms, we are introducing platform hooks. Platform hooks are a set of directories inside the application bundle that you can populate with scripts. These scripts are executed at defined points in the on-instance application deployment lifecycle. These hooks are reminiscent of custom platform hooks, but are simplified and easier to manage and version because they are part of the application bundle.

For example, a Corretto application bundle might look like:

├── .platform
│   ├── hooks
│   │   ├── prebuild
│   │   │   ├── 01_set_secrets.sh
│   │   │   └── 10_install_dependencies.sh
│   │   └── predeploy
│   │       └── 01_configure_corretto.sh
│   │   ├── postdeploy
│   │   │   └── 99_log_deployment_complete.pay
│   └── nginx
│       └── conf.d
│           └── custom.conf
├── Procfile
└── application.jar

The files in each of the .platform/hooks/ subdirectories are executed in lexicographical order at predefined points in the deployment process.

1. prebuild hooks are executed after the application is downloaded and extracted, but before we try to configure anything
2. predeploy hooks are run after the application is configured and staged, but before it is deployed.
3. postdeploy hooks are run at the very end — after the application is deployed and running.

Finally, take note of the .platform/nginx/ directory as well. This can be used to provide custom configuration additions or overrides for the on-instance NGINX proxy server. You can either override the provided configuration file completely, or just add a new configuration file that is imported by NGINX. Because all of the AL2 platforms use NGINX and the same base configuration, these customizations are now more portable across platforms. For a full explanation of platform hooks and related functionality, see our Extending Linux Platforms documentation page.

We’re excited to launch this new generation of Elastic Beanstalk platforms, and to hear feedback from you about how we can make them even better. If you have feedback about one of the AL2 beta platforms, please add a comment to the relevant issue on the public roadmap on GitHub. For example, here is the issue for the Corretto platform. Keep an eye on the roadmap and our release notes for announcements of the remaining platforms over the coming weeks.