Kubernetes — Container Orchestration from Google

Kubernetes was announced as an open-source project on 10 June 2014. That day Eric Brewer, Google's vice president of infrastructure, introduced it in his keynote on the second day of the first DockerCon in San Francisco, and in a post on the Google Cloud Platform blog. The repository's first commit, by Joe Beda, is dated 6 June 2014; the design work had started in the autumn of 2013 with Craig McLuckie, Joe Beda and Brendan Burns. Kubernetes 1.0 was declared on 21 July 2015 at OSCON, where Google also announced it would hand the project to the newly formed Cloud Native Computing Foundation (CNCF).

The name is Greek for the helmsman of a ship — Brewer's post spelled out the pronunciation, koo-ber-nay'-tace — which is why the logo is a ship's wheel.

The problem it was built for

Docker, unveiled in March 2013, had made it easy to package an application as a container and run it on one machine. It said nothing about running thousands of containers across hundreds of machines: deciding where each one goes, restarting it when it dies, keeping a set number running, and letting them find each other.

Google had been doing that internally for years. Brewer's post opened with the claim that everything at Google, from Search to Gmail, ran in a Linux container, and that Google launched more than two billion container instances a week. Kubernetes was offered as a lean, open-source container manager for developers whose needs were more modest than Google's.

Borg, Omega and Kubernetes

Kubernetes is often described as an open-source version of Google's internal system Borg. It is not a copy, and the 2014 announcement did not mention Borg at all. Brewer's post named Omega as the system Google used internally.

Borg became public the following year. In April 2015 Google described it in a paper for the EuroSys conference: a cluster manager running hundreds of thousands of jobs, from many thousands of applications, across clusters of up to tens of thousands of machines. The Kubernetes blog published a companion post on 23 April 2015 that explained the relationship: many of the developers working on Kubernetes had previously worked on Borg, and they had tried to carry over Borg's best ideas while fixing problems that Borg's users had pointed out over the years.

Autumn 2013 to June 2014

In a 2018 history for the Kubernetes blog, Brendan Burns wrote that the project began in the autumn of 2013, when he, Craig McLuckie and Joe Beda were working on public cloud infrastructure at Google. He recalled that it was probably Joe Beda who first suggested looking at Docker, in the summer of 2013. Its implications were obvious to all three — and so was the limit: Docker, with its focus on a single machine, was not the complete solution. The decision to build an open-source orchestrator crystallised in the late autumn of 2013.

As 2013 turned into 2014, Ville Aikas, Tim Hockin, Dawn Chen, Brian Grant and Daniel Smith joined. Burns described the goal as a "minimally viable orchestrator", and listed what that meant:

• Replication — keep a given number of copies running
• Load balancing and service discovery — let containers find and reach one another
• Health checking and repair — notice failures and replace what failed
• Scheduling — decide which machine runs which container

That, he wrote, was the sum total of the feature set released in June 2014. Part of the work before then had been persuading Google's leadership that open-sourcing the project was a good idea. The first commit on GitHub, dated 6 June 2014 and titled simply "First commit", added more than 47,000 lines. Four days later the project was public.

1.0 and the foundation

At OSCON on 21 July 2015, Google declared Kubernetes ready for production and announced that it would donate the technology to the Cloud Native Computing Foundation, a new body under the Linux Foundation. The founding members included IBM, Red Hat, Docker, CoreOS, Intel, Cisco, VMware, Twitter and Huawei, among others.

Burns wrote that the team spent the year after launch building, polishing and fixing that minimal orchestrator into 1.0, and credited the OpenShift team, which joined early, with much of the engineering and enterprise experience. On 6 March 2018 the CNCF announced that Kubernetes was the first of its projects to graduate from incubation, the foundation's highest maturity level.

What it became

Kubernetes arrived in a crowded field. Docker's own retrospective, written for the tenth anniversary of Brewer's keynote, recalls Apache Mesos, Red Hat's GearD, Docker's libswarm and Facebook's Tupperware appearing around the same time, and says that by 2017 the community had settled on Kubernetes for orchestration and Docker for the container format and runtime. That is one participant's account of the outcome.

The same year Kubernetes appeared, AWS took a different path with Lambda, which hid containers and machines behind single functions. The two models — orchestrated containers and serverless functions — still sit side by side.

For the concept, see Kubernetes and container in the glossary. The company is covered on Google, and the sequence on the cloud computing timeline.