Proactive not reactive. API-driven rather than scripts and SSH. Immutable, not repeatedly patched and reconfigured… When you’re designing cloud-native workloads to run on Kubernetes, you want them to be as scalable, automatable and (ideally) stateless as possible, the way Kubernetes itself is. Shouldn’t the same be true of the Linux that Kubernetes runs on?
Traditional Linux distributions like Red Hat and Ubuntu are mature and reliable; but critics say they weren’t designed with ephemeral workloads, containers and scaleout in mind.
Talos Linux rethinks the traditional approach to take some of the grind out of adopting and operating Kubernetes.
In many ways, Talos is trying to be invisible.
“Our entire reason for being is we want you to forget there's an operating system under Kubernetes, and we want to give you Kubernetes, period,” explained Spencer Smith, senior director of operations at Sidero Labs.
(Quoted in the article? Talos or Linux committer? Pop us an email and we'll give you a month's free subscription.)
Pare down not patch
The idea came from the “horrendous” auditing process that Smith and Talos creator (and founder of Sidero Labs) Andrew Rynhard went through at Dell when they were running Kubernetes clusters with Kubespray and Ubuntu, with all its packages and dependencies.
“We’re going through and proving we've patched this version of Apache, and we're not using Apache at all, but it's there on the box.”
Despite their expertise (Smith was a Kubespray maintainer), patching and updating and auditing over and over again was exhausting and didn’t actually address the underlying problems. “I can’t image operating Kubernetes that way any more.”
