Mesosphere DC/OS

» see dcos.io as it lived, on the wayback machine

~ the goodbye ~

I remember this from the window where you had to pick one. Roughly 2016 to 2018, if you were putting containers into production, the choice was Kubernetes or Mesos, with Docker Swarm still in the room, and it genuinely was not obvious which one would still be there in five years. I picked the one everyone else picked, for the reason everyone else picked it, which is that the tooling around it was moving faster than the alternatives.

Mesosphere was founded in 2013 by Florian Leibert and Tobias Knaup, joined by Benjamin Hindman, who had co-created Apache Mesos at Berkeley. The pitch was cleaner than anything Kubernetes was saying at the time: pool every machine in your datacenter and treat the whole thing as one computer that you submit work to. DC/OS was the commercial packaging of that idea, sold for months before it was open sourced on 19 April 2016 alongside sixty launch partners.

The customer list was not marketing. Mesos ran all of Apple's Siri. It ran large parts of Airbnb and Netflix, and Verizon moved onto DC/OS in production. For a technology that ended up as a footnote, it spent its peak years underneath things hundreds of millions of people touched daily without knowing the name.

~ the chronicle ~

~ the breakout ~

Investors put close to $126 million into Mesosphere across fourteen backers, most of it in the stretch when "datacenter operating system" was a phrase people said seriously in keynotes. The company shipped a package catalog and a load balancer around the Marathon scheduler, which a lot of shops ran on bare Mesos without ever paying for the rest.

~ the decline ~

The pivot started early and it was visible from outside. DC/OS 1.11 shipped Kubernetes support in 2018, which meant the datacenter operating system was now a way to run its own competitor. By 1.12 the marketing was about running multiple Kubernetes clusters. On 5 August 2019 the company changed its name to D2iQ, and CEO Mike Fey said plainly that the Mesosphere name had become constrictive because it tied them to a single technology.

Underneath the branding, the practical complaints were mundane. Mesos was harder to run than Kubernetes because writing a framework correctly was hard. DC/OS did not ship a cron job runner out of the box. None of that is fatal on its own, but every gap had to be closed by one company's engineers, while the other side had the entire industry filling in its gaps for free.

~ the inheritance ~

The end-of-life notice went up at the end of October 2020 and set the date exactly one year out: 31 October 2021. D2iQ said Kubernetes had reached the capability that only DC/OS used to have and was improving faster, which is an unusually honest thing for a vendor to write about the product it is switching off. Existing contracts were honored, new features stopped, and the repositories stopped being maintained.

D2iQ itself did not last much longer as an independent company, and today d2iq.com redirects to a Nutanix page. Apache Mesos survived that round by two days: a vote on 5 April 2021 to move it to the Apache Attic was canceled when enough remaining users volunteered to keep it going. On a Hacker News thread about the shutdown, one commenter wrote that it was sad to be a technologist who works on something for years only to watch it become irrelevant. That is the actual epitaph and nobody at the company wrote it.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »