Future of Kubernetes configuration, profile-guided optimizations and AI

Bart: Who are you? What's your role? And who do you work for?

Frederic: My name is Frederic. I founded a company called Polar Signals, and we do continuous profiling.

Bart: What are three Kubernetes emerging tools that you're keeping an eye on?

Frederic: I think I'm maybe one of the worst people to ask this because I'm one of the last people to adopt new tools. I stay away from tools for a very long time. But if I had to pick three, one that I'm super excited about is GPT-Script. I don't know if it's really that much of Not that much Kubernetes related, but basically the folks at Acorn Labs have created this thing where you can create tools that call real code through AI. And it's pretty cool where you can kind of mix the world of AI and code. It's really awesome. Then the next one, I'm super excited about people exploring packaging further. So I think the Timoni project is pretty cool, trying to use Cue for packaging. Maintaining Helm packages is pretty painful. So I'm excited that people are trying out new ways that hopefully won't be as difficult. And then actually the folks at Authzed have just created a really cool proxy that you can put in front of your Kubernetes cluster, which basically makes it multi-tenant, but in the most awesome of ways. And I would have to talk a lot more about it, but check it out. I think it's really cool.

Bart: One of our guests, Jacco, dislikes Helm's approach to templating, mentioning difficulties with multi-line strings and loss of strict schema enforcement but acknowledged the usefulness of Helm packages. Do you see more tools and companies tackling the Helm Package Manager with new and innovative solutions? Are you happy with the status quo? How should we install third-party packages into Kubernetes?

Frederic: So, I spend a very large amount of time on patch packaging. One of the most widely used ways to deploy Prometheus on Kubernetes is called Kube Prometheus. We wrote all of that in JSON, which a lot of people don't have a lot of love for. One of the things that I really love about Jsonnet is that you can arbitrarily change and merge patches on top of Jsonnet. And there's really no other language out there that can do something like this. I am excited about Cue. I think it's, in my opinion, the next best thing. And I think Cue is going to probably catch on more because I think the developer experience is just so much better. And so, I think that's where things are going.

Bart: One of our guests, Alex, spent several weeks troubleshooting an issue with Kubernetes, which took the team to explore the kernel code. He stressed the importance of learning while troubleshooting. Is there any practical advice that you've learned during the years when it comes to debugging?

Frederic: I wasn't actually planning on plugging our own product, but basically, our product can show you what code is running on the CPU, no matter what code it is. It can be your user space code, or it can be kernel code. It gives you truly whatever code is running on the CPU, it's going to show you. So, I actually think this is super useful. We run these weekly live streams that we call Let's Profile, where we take a lot of the open source projects that are created, maintained, and shown up here, and try to optimize those. We show exactly those kinds of things.

Bart: Kubernetes is turning 10 years old this year. What should we expect in the next 10 years to come?

Frederic: Well, someone once told me that everything that's six months or older is probably not going to happen in any way, shape, or form that you imagine. I do obviously think AI is going to play a big role, but I think it's going to be more subtle than we probably think. I think it's going to augment us and guide us more than it does everything for us, making us significantly more productive.

Bart: What's next for you?

Frederic: For us at Polar Signals, something that's really exciting is we're basically turning profiling data into change. We've built a really awesome profiling product that can profile just about any language. And now we want to turn that into real improvements. One of the really exciting things in this ecosystem is something called profile-guided optimizations. It's something that the Go compiler recently introduced, where you can feed profiling data into the Go compiler, and it will apply optimizations that it wouldn't usually apply. But now that it has information about how the code is actually executed in production, it can apply optimizations in much better ways and therefore make software faster without having to change any code. It's super exciting.

Bart: How can people get in touch with you?

Frederic: So we're on Twitter, obviously, or X, as people call it now. Like I said, we run this weekly live stream on YouTube at PolarSignals.io.