I think everything is wrong with Kubernetes


Kubernetes is cool, but that’s not why you’re thinking. For a while, people noticed that Kubernetes was promised to be a great new cloud technology like OpenStack (everything was fine). But it wasn’t Kubernetes. Nor was it the magic cure for lock-in to ensure unlimited portability. far cry.

Instead, as I wrote, Kubernetes is the new Linux. Or more precisely, this is a new app server, as Alexis Richardson, CEO of Weaveworks, suggested in an interview. Instead of companies trying to build their own cloud, he argues, “Let the development team run a short-lived Kubernetes cluster like an app server.”

What exactly does that mean?

Technobabble and Gibberish

Richardson and I should be talking about multi-cluster Kubernetes on bare metal with a microVM with Firecracker. After 5 seconds my eyes were dazzling and I was driving so it was a problem. “In addition to adapting to mixed-mode edge scenarios, moving control plane nodes from a dedicated bare metal server to a microVM improves the management efficiency of bare metal Kubernetes clusters and the overall number of nodes required for a bare metal pool considerably. Is significantly reduced. He cites the company’s blog post on the matter to set the tone.

Given that my employer released Firecracker as an open source project, I desperately tried to be careful. But I could not. I have not been able to find exciting telecom companies delivering 5G applications that can perform “network functions” (5G workloads) with signaling, management, web and customer applications on the same hardware with strong isolation and resource control.

Somebody takes care of it. I did not do it.

At that point, Richardson said, are you there? Hello? “And I thought about hanging up and pretending the signal was down, but it’s all” a super efficient way to start a fleet of Kubernetes clusters with minimal resources, “he said Remove glaze just a little.

Kubernetes as an app server

when Really However, it makes sense if he compares Kubernetes to Linux or an app server. This is a world that I knew and grew up in technology during the JBoss / BEA era of app servers. An app server is a collection of generalized functions that allow you to run an application by embedding it in the application or by running it first and then running it. I won’t talk much about them anymore, but it’s very common in businesses.

In the app server model, the life cycle of an application is mapped to the life cycle of the stack or application cluster in which it runs. When you no longer need your infrastructure, just shut it down and destroy it. The clouds are different. It is persistent even if the running resource is not persistent. The idea of ​​the cloud is that someone takes care of the cloud and survives beyond the requirements to use it. Rather than companies building their own clouds, Richardson said, “We’re going to have Amazon, Google, and Microsoft run the cloud, and let the development team run short-lived Kubernetes clusters like app servers.”

Richardson’s thinking of Kubernetes as an app server is a powerful principle for corporate IT. Release the development team. “In practice, it is much better to allow individual application teams to set up their own infrastructure using Kubernetes clusters than to pull enterprise cluster resources from central IT,” he said. Says. Encourage “single use applications, groups of applications that can be managed by a small number of groups, or groups of applications that can be managed by one person” rather than large Kubernetes clusters where individual development teams must be borrowed. It is recommended. The best practice is to have a Kubernetes cluster and a large number of Kubernetes-supported applications instead of trying to manage it as a cloud. “

This is at odds with what Kubernetes believes are complex and difficult to set up. Maybe. The difficulty, however, is often how to treat Kubernetes like a centralized cloud instead of enabling features like an app server in a small team. It’s about autonomy, explains Richardson:

“Development teams are more productive when they are allowed to own a business. You need to know that if something goes wrong, they can solve problems. Otherwise they are in safety and performance. We cannot take responsibility. We are expanding it with Weaveworks. Kubernetes and GitOps are absolutely free so that development teams can start and stop infrastructure, clusters, and stacks. Has helped. You can get it on request. In the past, developers could start and stop Tomcat. You are not asking anyone’s permission to do this. That’s what Kubernetes is all about. It works the same way for companies. “

This is an incredibly liberating vision for modern IT. This is also a great way to understand the technobabble that started the conversation. “The blog post I shared with you just goes to show that this can be done more efficiently with a wide variety of use cases, including edge workloads,” he says. Well, why didn’t he say that at all?

Copyright © 2021 IDG Communications, Inc.


Comments are closed.