Shaun O’Meara, Mirantis | KubeCon + CloudNativeCon North America 2023

This is Textron tv. Hey, everybody. Welcome back to KU Con 2023 North America, here in Chicago.

So we are talking Kubernetes, Kubernetes, Kubernetes. There's some other topics to, that's certainly a theme for today. So I'm joined by Sean o' Welcome Sean from ranches.

Thanks For having me. A field CTO, correct. Field, CTO.

That's Right. Yeah. Great.

Say a little bit about Martis, the folks may not know, know, know of you all and kind of things that you do, and we'll talk about some of your announcements. Cool. Uh, Martis been around for a better part of 12, 13 years now.

Very focused in the open source space. Um, specifically we started off in the infrastructure space with the OpenStack community. Mm-Hmm.

Moved to Kubernetes fairly early, um, you know, collaborated with Google in the early days on Kubernetes. And we've really just focused on enterprise customers. Uh, in 2019, we bought the Docker enterprise business, which we've merged into our product line, um, Marant Kubernetes engine product line.

Um, and, you know, we're running hundreds of customers on Kubernetes today and have done so for at scale for a number of years. So our entire business is around how to make Kubernetes easier and more consumable. A lot of depth text and research, which I run.

And we've done some work with you all, and on that, some great, uh, interesting research. So, um, what's new? Kind of what's happening now?

You're, you're, you all continue to move and move fast and Yep. Kind of do big things when you, when you tend to announce things. Yeah, We, we like to make a splash when we can.

Um, you know, we really needed to double down on our open source and our open source chops and everything we've been doing in that space. Um, so round the 2021 timeframe, we launched our K Zeros offering, which for us was the next generation of the Kubernetes offering. Um, really focusing on how do we make Kubernetes super easy to use, um, to deploy.

But the challenge with K Zeros, as easy as it is, it's still one cluster. The majority of our customer base today though, run one or two large clusters. And that is a management nightmare for them.

Um, because you gotta learn Kubernetes, you gotta manage that. So just before the show, at the show, we've just announced Cosmo Tron. Cosmo Tron is a new way of deploying and managing Kubernetes clusters.

So we are really trying to change the paradigm of how people think of a Kubernetes cluster. Cluster can be a throwaway, it can be a, it doesn't need to be a pet, it can be cattle. Uh, you know, to use the old paradigm here, um, really what we are trying to solve for with, with Cosmic Tron is this idea that you can have a small cluster, very affordably.

That's the, the, the key message. Ultimately, what we're doing with Cosmo Tron is we're solving the problem of the resource overhead. When I need to build clusters, I want smaller clusters, but I don't want to deal with the control plane overhead, you know, the three to five nodes, and then a hundred clusters, a hundred nodes behind that cluster, which I then gotta manage and deal with all the complexity of if I can put Cosmo Tron onto my systems and I can make virtualize the control planes.

So run them as a Kubernetes workload and then attach workers from anywhere. So that's the key message here. What we've done is I can spin up a new Kubernetes cluster in about seven seconds, three lines of yaml, and there I go.

I have a cluster up and running. And then the other part of the, what we've done is we've integrated that to capi. So the cluster API service, um, made Cosmo tron a bootstrap provider for capi, so I can now attach workers from anywhere.

Yeah. So plus if you have existing infrastructure tied into that as well. Right.

It's gonna be my first question. So I New Greenfield, or can you bring it to So some, so you're talking about people having generally one large or couple large clusters, and Of course, yeah. Those things Mm-Hmm.

Probably should be other things, but who has the time to really kind of manage it, split it up, uh, in that kind of a scenario? Is it still, you know, attach, uh, Cosme tron to the workers and you can start managing that? Or is there sort of a dividing line of where very oriented towards this kind of a size of an implementation versus something that's bigger?

So, so ultimately what we're doing with Cosmic Tron is we're just moving the control plane from being a pet. So by virtualizing the control plane, I can now spin up hundreds of control planes. So instead of having to share a cluster or share workloads on a single cluster, I can spin up a cluster per workload.

Mm-Hmm. Because I don't have that overhead. Um, the way cosmic Tron works, we use the connectivity with AK service.

Um, so now we have full separation between those workers and that control plane. So now my workers can live anywhere Mm-Hmm. Um, so I can put the workers and I can run a Cosmo tron control plane on-prem.

I own my control plane. I don't have to use a third party control plane, you know, um, use one of the cloud provider control planes, which take 25 minutes to spin up. But I can attach workers from any of those cloud providers or all of those cloud providers as long as I have a connectivity back to that single control plane.

Mm-Hmm. Um, suddenly I'm changing the way, if I want to do testing work, um, I spin up a cluster, a whole cluster for that test, run the test and kill it. So that's the beauty of what's behind Cosme Control.

It seems like almost you've, uh, abstracted or kind of, uh, virtualized away the structural component of Kubernetes and said, that's still there. Right. You, you need those things for, for Still performance, Have lots of good reasons, but I can also now kinda treat it as anywhere.

Right. Matter of fact, it seems like those one or two big cluster implementations would now you could start spinning up wherever you want. Right.

Let's new cloud provider on-prem, maybe same cloud provider, but True multi-cloud cluster Two, true multi-cloud. Yeah. Because instead of trying to force my developers to use a bunch of different ed points and complicate the application deployment, I'm now doing that and I can put multiple qubits in different cloud locations or at the edge attached to that centralized control plane.

And I'm deploying to one, one single cube API. Trying to just simplify, again, with our whole processes, simplify the developer experience of getting actual workload into production, which is what every company ultimately wants. Seems also like it would help, uh, kind of the development process too, right.

Early development, development environment, testing environments, simulating production. Yeah. CICD processes, you know, being able to spin up clusters on demand, being able to move your production workload really simply.

You know, if your application is fully automated, why shouldn't your infrastructure be fully automated? Good point. Um, so Talk some more about the, the control plane.

And you said you containerized it. Uh, how does that differentiate from its imple the way it's implemented now? Is that one of the limitations of expanding and growing and managing across a complex infrastructure?

So It all really boils down to the Cub A-T-A-P-I and et TD. Um, typically when I'm deploying a control plane, I need to deploy ET TD and I need to deploy the cube, API, and they're, they're distinct components of the system. So if I'm using something like Cube, ADM to deploy a control plane, I've gotta spin up three separate nodes, at least four production.

I mean, yes, you can do a single node cluster, but it's not really a cluster. Uh, what I'm doing is I'm, I'm putting ET CD and the cube, API into a single pod, and I'm treating those as just another Kubernetes containerized workload, which means I control them using Kubernetes, Kubernetes norms and Kubernetes standards for containers. So ku, API et CD lives inside the pod.

I can make as many replicas of that pod, or not as I want to change a replica count in Kubernetes standard CRD norms for Kubernetes beauty is now I need to expand that control plane to support more workers. I push a button two seconds later, I have another, another pod running or another version running, another container running. So I'm expanding.

Um, the cubelets itself runs anywhere, and then the connectivity service provides for that connectivity. Great. You, you mentioned emphasizing open source kind of really, I guess kinda leaning into that even more.

Right. Tell me about some things you've done. Is that particularly around Coss tron or is it in general, or what, what's the philosophy There?

It's pretty much in, in general, you know, the lens team that, that where this work has come out of the lens group, um, we really wanted to focus on, on, on open source as much as possible. We, we leverage the open source community for everything we do at tis. We've been huge contributors to the community over the years.

We, we want to continue to be good citizens of the open source community. So both K Zeros is fully open source. Um, and we've got a huge following of K Zeros and a lot of people providing input and feedback.

We just move faster in the open source space. We can do things with much better quality, um, because we get that feedback in a very controlled, and so both K Zeros and Cosmo tron are fully open source. We obviously offer commercial support options for them.

Um, but we're seeing a lot of customers are diving in there, trying it out, implementing it, and then coming to talk to us later on about getting support for it just to get that backup So you, it takes you outta the game of kind of the cripple wear. Right. Of it sort of works.

But if you need more than five, yeah. You gotta come to us. Right.

So, you know, ultimately the open source model is always challenging for any, any company. Um, we want to put it out there, but we also need to pay for it to be produced. Um, so by being good citizens, people can consume our software when they go into production, um, they need that extra layer of support or quite frankly, they don't have the people to do that support themselves.

Yeah, I was thinking that too. And that's who really we're talking to. We're talking to a lot of platform teams who want to focus their efforts on building the tools their developers need to be more efficient.

Well, we are providing the tools, the platform teams need to be more efficient. Um, and if they can treat Kubernetes as just another workload, same way they treat those workloads, we are simplifying their lives and giving them back time to focus on building value for their business. That's ethos.

It's kind of built into the nation of the job. But, you know, organizations aren't hiring 5200, 200 platform engineers. They've got 1, 2, 3, 4, 8, maybe, you know, not not large teams.

So they're kind of incented to do their own productivity improvement by open source automation, you know, management control plane Kind of thing. Absolutely. I mean, you know, I, I was, I've been going around the world talking to a lot of the customers.

Uh, I was in the Nordics recently with a platform team at a bank. Um, and they're super excited by what this brings them because they're spending so much of their time dealing with platform underlying platform issues. Mm-Hmm.

At what is the commodity layer? You know, the Kubernetes API is really a commodity layer. It's what we put on top of it that brings value to the business.

And I can get Kubernetes now from, you know, half a dozen to a dozen different places. Yeah, absolutely. On demand or deploy it myself.

It, but it's still complicated and it shouldn't be. And that's what we're trying to, trying to move for the community to be into that place. Is There, is some of that complexity also introduced by this dis these distributions or these cloud providers?

Or especially you talked about multi-cloud does that now you're talking about not everything's not uniform the same. I've gotta manage across that variation. I mean, of course it is.

The ultimately, you know, a business model says you want to try and keep your customers as close to home as possible. And I wouldn't point fingers here to anybody, but the lockin model is a true thing. Um, especially in our European customers, we're seeing more and more desire to use multiple providers.

And I'm being very careful to say providers. 'cause that could be cloud or on-premise, right? Right.

It could be, often we're seeing two vendor policies coming in at least two vendor policies. So they'll work with us and another vendor or you know, two other vendors providing their Kubernetes resources. The reality of that is it's not just the Kubernetes that's different, and often it's not, you know, at least at the surface, the AP is the same.

It's everything that they wrap around that, that is different and adds layers of complexity for platform teams as well as, as I mentioned earlier, the multiple endpoint problem. You know, I'm deploying my app now. Where do I, which endpoints do I choose to?

And suddenly what ends up happening is the apps just go into one, one provider 'cause it's easier. Um, so we're seeing that, that the time saving and the simplicity that we can offer is really good for these, these platform teams. Great.

If you don't mind, I'd love to ask about K Zero. Is that primarily being used, um, kind of spin up those simplified development environment or do you see it being used at the edge a lot? 'cause people talk about K three K kind of edge environments?

Yeah, so we've got a wide range of K Zeros customers today and a wide range of user use cases. Um, currently I think we've got about 200,000, um, clusters that are, that are reporting telemetry. Um, but we see that we, we've got use cases right now with people building clusters on-Prem.

So traditional, you know, expand our clusters. We seeing clusters that are being built for in the public cloud. So on top of public cloud providers, on top of VM instances, we're seeing it being used as appliances.

In fact, we have one customer who is embedding K Zeros as a single node, single cluster in invert to commas appliance. And they push K zeros out with the K Zeros manifest capabilities and they run it as an appliance. We're seeing K Zeros being used at the edge.

Um, 'cause the cubits is so simple to deploy. We just pop out, pop it out, provide a, provide a token file and it can run, you know, I hate to say it on something like a raspberry pie or an equivalent commercial version of a raspberry pie. Mm-Hmm.

So, really makes it, it it, it's been designed to be simple to use in a multitude of different use cases. And Edge is definitely one of those, It seems kind of architecturally that really supports the idea of, you know, cars are becoming software platforms with wheels on it. Right.

For sure. But everything is like your smartphone could be some part of that distributed edge network for you, or it Already is in some ways. Right.

And that seems like a great place where K Zero could be. Absolutely. I mean, it really could permeate, permeate a lot of environments.

Well, the beauty of that is, you know, we're looking at it from the point of view of K Zeros being the simple to deploy and manage Kubernetes Cosmo Tron, removing the complexity of the control plane kpi, providing that infrastructure management layer. Um, and we've done some extra work to, to do bare metal management and infra, um, operating system management as part of one of our other products. All of those things combined, I now have a very simple way to do a very, very distributed Kubernetes network.

Um, and because I have all this information in a central database, so from an application point of view, I can query that database and understand what my left work looks like. Mm-Hmm. And that's the beauty behind it.

Great. capi, Cosme, Tron, and K Zeros, all of that together. And that's the vision.

So you launched, uh, Cosme Tron here. Have you been having open source users of that for a while now? I assume it's been out there and kind of tested and battle tested?

So we've been doing demos here all week. Um, we have a number of people trying it out. Um, you know, we're seeing like K Zeros.

We've got a lot of commercial, commercial use. Cosmo tron is is new. Um, we've got a lot of test clusters out there, a lot of cappy integration work happening.

Um, I've been doing demos here all week showing people, you know, three lines of yaml, seven seconds, and there's a cluster up and running. Mm-Hmm. Um, I can do a full CAPI deployment on a public cloud provider in under a minute and a half, and most of that time is waiting for the instances to start.

Mm-Hmm. So it's, it's pretty stable. Um, we're seeing, we're seeing a lot of good uptake.

Uh, we really want the community to come out and dig into it and provide us feedback. Um, we're always trying to improve. That's the open source model.

Mm-Hmm. Um, where can folks find out more, kick the tires, download open source, do all the good things you can Do? Everything's on GitHub.

com. There are links to all of this. Uh, we are doing everything in the open.

The team is super active online. Um, so, you know, dig in on GitHub, ask questions, load issues. The guys are always there to respond.

Um, the leadership of the, the K Zeros team has been helping me get ready for this preparing videos and demos at two o'clock in the morning their time. And they're all in Finland. So it's been a great, great experience to work With them.

You're markedly uh, you know, vivid for being, being up that long, doing those long hours of work. So No, great team guys for doing Well. Well, congrats on the launch.

Thank you. And the hard work on, uh, Cosme Tron and K zero and other great things tis is doing. So keep, keep us informed.

Let us know how it's going. And, uh, you're really interesting to see where this evolves and use cases that Mike has. It's always the fun part.

It's always the fun part. We never intended that to happen, but it's a great use case, Right? Yeah.

We intend to publish those use cases, um, as soon as they're, as soon as they're ready. Um, we love talking about what we do. Uh, we put a lot of our use cases on the websites as well.

So we, we are doing some amazing stuff with customers around the world, so. Good. Good.

All right. Sean Oir here from Meranti. Be sure and check it out.

I'm GitHub and their website. We'll be back with another great inter interview here from Kon North America in Chicago on our third day of live streaming, uh, about Kubernetes and a lot of other great inning. So thank you Sean, For coming.

Thank you very much. We'll be back in just a minute.