rkt is a new container runtime, designed for composability, security, and speed. Today we are releasing a prototype version on GitHub to begin gathering feedback from our community and explain why we are building rkt.
When we started building CoreOS, we looked at all the various components available to us, re-using the best tools, and building the ones that did not exist. We believe strongly in the Unix philosophy: tools should be independently useful, but have clean integration points. We hope this is reflected in tools that we build, such as etcd, which have seen widespread adoption and use outside CoreOS itself.
When Docker was first introduced to us in early 2013, the idea of a “standard container” was striking and immediately attractive: a simple component, a composable unit, that could be used in a variety of systems. The Docker repository included a manifesto of what a standard container should be. This was a rally cry to the industry, and we quickly followed. Brandon Philips, co-founder/CTO of CoreOS, became a top Docker contributor, and now serves on the Docker governance board. CoreOS is one of the most widely used platforms for Docker containers, and ships releases to the community hours after they happen upstream. We thought Docker would become a simple unit that we can all agree on.
Unfortunately, a simple re-usable component is not how things are playing out. Docker now is building tools for launching cloud servers, systems for clustering, and a wide range of functions: building images, running images, uploading, downloading, and eventually even overlay networking, all compiled into one monolithic binary running primarily as root on your server. The standard container manifesto was removed. We should stop talking about Docker containers, and start talking about the Docker Platform. It is not becoming the simple composable building block we had envisioned.
We still believe in the original premise of containers that Docker introduced, so we are doing something about it. While we are at it, we are cleaning up and fixing a few things that we’d like to see in a production ready container. What is important to us in the design of a container?
rkt is a command line tool,
rkt, for running App Containers. An “App Container” is the specification of an image format, container runtime, and a discovery mechanism. rkt is the first implementation of an App Container, but we do not expect it to be the only one.
The best way to drive a standard is to let a successful and proven implementation become the de facto one. However, when the point of the software is interoperability (as it is with containers), we think things are different.
In developing the App Container specification, we started with some thoughtfulness around the requirements up front, then spent time refining it as we worked on an implementation (rkt). We feel these specs and implementations are pretty well thought out, but are still early enough that we need your help refining it. Please contribute to the spec by sending a pull request to start the discussion.
An App Container Image (ACI) is a specification for the image format of a container. It is a simple flat tarball that is always signed and optionally encrypted. By convention, an ACI is minimal, meaning it only includes the bits absolutely required to execute the application. Since an ACI may be encrypted, distribution via systems like BitTorrent, public object storage, or mirror networks is a possibility. Think of ACI as a variant of Amazon’s AMI, but created for a container world.
The App Container Runtime defines what environment and facilities a container runtime should provide. This includes devices, environment variables, and privileges that a container should expect. It also includes a definition of a meta-data service interface for exposing data to the environment from outside the container.
Security primitives are very important to us, so we added an identity feature to the meta-data service. This means every instance of a running container is given a unique identity, coupled with a lightweight HSM-like service for signing. No existing VM or container environment has a concept like this, so we would welcome and appreciate community feedback on this design.
App Container Discovery is a method for finding your app container image. It is inspired by golang’s vanity URL convention for import paths. This means you’ll be able to refer to containers with simple names like
coreos.com/etcd, allowing organizations to federate their downloads without running their own registry.
We want the world to run containers. A world where your application can be packaged once, and ran in the environment you choose. We believe rkt and a definition around the App Container is a requirement for this to work. Please contribute your thoughts to our mailing list, our source code or join us in person at the CoreOS meet-up tonight in San Francisco.
rkt is an alternative to the Docker runtime, designed for server environments with the most rigorous security and production requirements. rkt is oriented around the App Container specification, a new set of simple and open specifications for a portable container format.
rkt is available today on GitHub. We are releasing a 0.1.0 prototype to gather community feedback. Please note that this is a prototype quality release, very much in the spirit of “release early, release often”. Please provide feedback via GitHub.
From a security and composability perspective, the Docker process model - where everything runs through a central daemon - is fundamentally flawed. To “fix” Docker would essentially mean a rewrite of the project, while inheriting all the baggage of the existing implementation.
At CoreOS we have large, serious users running in enterprise environments. We cannot in good faith continue to support Docker’s broken security model without addressing these issues. Additionally, in the past few weeks Docker has demonstrated that it is on a path to include many facilities beyond basic container management, turning it into a complex platform. Our primary users have existing platforms that they want to integrate containers with. We need to fill the gap for companies that just want a way to securely and portably run a container.
rkt is a stand alone tool that will live outside of CoreOS itself and can be used on a variety of platforms. In a sense it is similar to the etcd project in that it is a tool we built because nothing like it existed.
“App Container” defines a specification of the facilities surrounding the container. rkt implements these facilities as a command line tool. An open specification allows other systems do their own implementation of App Container without using rkt at all. CoreOS fully supports and embraces alternative implementations.
Definitely. If the App Container specifications were implemented inside of Docker, the projects will be interoperable, meeting the original goal of the manifesto. CoreOS will evaluate contributing this work once App Container matures.
Yes. We will continue to make sure CoreOS is the best place to run Docker. We will save the details for a future post, once rkt has developed further, but expect Docker to continue to be fully integrated with CoreOS as it is today.