API Transition Timeline

Once upon a time, on a Thursday afternoon somewhere in Italy, a KernelCI backend API was created:

commit 08c9b0879ebe81463e124308192670c0e7447e0b
Author: Milo Casagrande <milo@ubuntu.com>
Date:   Thu Feb 20 16:10:41 2014 +0100

    First commit.

As you can see, it was nearly 10 years ago. How much does that represent in the modern software world? Of course, it depends. The Linux kernel is much older, still written in C and still going strong. But in most cases, including this particular one, it means a whole new world. The old API was written in Python 2.7 which stopped being maintained as a language on 1st January 2020. We could have just rewritten it in Python 3, which was the initial thought. But in the meantime, KernelCI was also growing as a project. It wasn’t just about building ARM kernels and doing boot testing on embedded dev boards any more. It had become a Linux Foundation project aiming to test the whole upstream kernel.

What is this new API?

Following this move, an increasing number of people became interested as it got under the spotlight. That is when we started to realise that the architecture needed to fundamentally evolve in order to match the scale of the new mission it had been assigned. The Linux kernel is a vast and complex open-source project with a unique ecosystem. As such, it requires some unique tooling too. We all know that Git was initially created out of a need to manage all the kernel patches. Now KernelCI needs an automated testing tool tailor-made to its unique requirements - and that’s why we’re finally launching the new API & Pipeline.

It comes with lots of improvements and it’s still a work-in-progress. We’ll keep publishing blog posts and update the documentation as things evolve over the next few months. Right now we have a pipeline that can monitor Git repositories for new revisions, take a snapshot of the kernel source tree in a tarball, run KUnit with it as well as an x86 kernel build and smoke test it in QEMU. It can then also send a summary email and detect regressions. That’s basically enough to prove we have a workable system. Nothing too groundbreaking there, you might think. So, what’s all the fuss about?

In a nutshell: a Pub/Sub interface to orchestrate distributed client-side services that can be run anywhere. You could have your own too at home. Also: user accounts so you can keep your own personal test data there, an abstraction for runtime environments so jobs can be run seamlessly in Docker, Kubernetes, a local shell, LAVA, [insert your own system here]… a new kci command line tool to rule them all and a unified Node schema to contain all the test data (revision, build, runtime test, regression…) in a tree. But again, we’ll go through all that later in more detail. It’s all based on requirements gathered from the community over the past few years.

Timeline

The main message in this blog post is the timeline for retiring the old system and getting the new API in production. Here’s the proposal:

Early Access

Monday 4th September 2023

Production Deployment

Monday 4th December 2023

Legacy System Deprecation

Monday 4th March 2024

Legacy System Sunset

Monday 4th November 2024

It only takes four Monday-the-Forth milestones to get through all this. Here’s what they mean:

Early Access

This is when a new API & Pipeline instance becomes available to let the public experiment with it. It can be seen as some form of beta-testing. It will be deployed in the Cloud to evaluate how a real production instance would behave, but it’s only kept online as a best effort. There should be frequent updates as the code evolves, probably at least weekly and at most daily. Only changes that made it through early testing on the staging instance should be deployed so it’s meant to be reasonably stable.

Production Deployment

The plan is to build upon the experience learned from the Early Access deployment to prepare a persistent instance that would eventually become the production one. Data should be carefully kept and backed up, changes in the database schema should go through managed migrations and the API code should be deployed from tagged releases. As soon as this has become reliable enough we might shut down the Early Access instance since it should have become redundant by then.

Legacy System Deprecation

In other words, this is when the new API & Pipeline production instance becomes the official main KernelCI instance. We’ll first be going through a transition phase to ramp up the build and test coverage on the new API while equally reducing the load on the legacy system to avoid overloading the shared infrastructure. Ideally, coverage should have reached 80% on the new API and 20% on the old one by this date.

Legacy System Sunset

After being deprecated, the legacy system will keep running with a bare minimal amount of coverage just to facilitate the transition for users who depended the most on it. It will be definitely shut down and the data will be archived when finally reaching the Sunset milestone.

Stay tuned

These dates have been identified as realistic targets for having the new API rolled out and retiring the old one with a transition in between. We’ll be aiming to have the new API in place by these dates and conversely retire the legacy system no sooner than announced here.

In the meantime, we’ll be posting updates as these milestones get reached or if any alterations need to be managed. We’ll also clarify how to use the API and exactly what features become available alongside the main documentation. Please share with us any feedback you may have, if you need some clarifications or to raise any concerns. The best way to do this is via the mailing list as always. Stay tuned!