← Kubernetes Podcast from Google
Kubernetes Podcast from Google
Kubernetes & Cloud Native Trends, with Alain Regnier and Camila Martins
00:31:21
This week’s interview was recorded live at Google Cloud Next, and features and talking about recent developments in Kubernetes and cloud-native technologies. Including exploring highlights from KubeCon EU, and the value of community events....
Loading summary
Transcript81 lines
- [00:00]
Kaslyn Fields
Hello and welcome to the Kubernetes podcast from Google. I'm your host Kaslyn Fields.
- [00:05]
Mofean
And I'm Mofean.
- [00:19]
Kaslyn Fields
This week's interview was recorded live at Google Cloud Next and features Alain Renier and Camilla Martins talking about recent developments in Kubernetes and cloud native technologies in including exploring highlights from Kubecon EU and the value of community events. But first let's get to the news.
- [00:40]
Mofean
AWS introduces specialized Model Content Protocol or MCP servers for Amazon, ecs, eks, FIN and AWS Serverless, providing real time contextual responses and service specific guidance to guide AI assisted application development.
- [00:54]
Kaslyn Fields
Summer is the season of Kubecon events in Asia. Kubecon China took place June 10th to 11th, Kubecon Japan was on June 16th and 17th and Kubecon India is coming up August 6th and 7. Be on the lookout for cool announcements coming out around these events.
- [01:13]
Mofean
A reminder for CNCF Open Source Project Maintainers the upcoming Maintainer Summit co located at Kubecon North America has an open cfp which closes July 20 and the maintainer Trap CFP for Kubecon North America closes on July 7th. These opportunities are only available to CNCF project maintainers, but if you are one, make sure you take advantage of these opportunities to share the awesome work you're doing with users and fellow maintainers.
- [01:40]
Kaslyn Fields
A new open source project LLMD aims to create a well lit path for serving large language models or LLMs on Kubernetes. It combines VLLM with the Kubernetes Inference gateway to enable smart state aware routing that dramatically improves latency for multi turn conversations by leveraging KV caching.
- [02:02]
Mofean
Speaking of Gateway API Inference Extension, a blog on Gateway API Inference extension was published. Inference Extension aims to make serving LLMs on Kubernetes easier with much needed features like model identity, aware routing and request criticality.
- [02:17]
Kaslyn Fields
And that's the news.
- [02:21]
Camilla Martins
Hello everyone, we're coming to you live from Google Cloud Next, where I'm very excited to be recording Kubernetes podcast episodes for video as well as audio. So if you're listening to this on audio, there's also a video that will probably be posted on our YouTube channel. And today I'm excited to be talking with Camilla and could you pronounce your name for me? I didn't get to ask you this one. Alain Alain, would you like to introduce yourselves?
- [02:49]
Alain Renier
Yes. Hello, I'm Alain Rinier. I'm a CTO of a small startup based in Paris called Cubo Labs and we develop tools to help identify and solve Issues on your Kubernetes cluster very quickly.
- [03:02]
Camilla Martins
Excellent.
- [03:02]
Yeah, My name is Camilla Martins. I'm from Brazil. I'm a GDE from Google Cloud and also Modern Architecture. I organize some DevOps Day's events like CNCF KCD events meetups, the KCD Rio de Janeiro actually and DevOps Days through Brazil and. And Le Tan. So yeah, I'm so involved with the community and also I'm a Senior SRE at StoryBlock too.
- [03:34]
Yeah, thank you both so much for being on today. I'm excited to talk about some of the things that you all have been doing recently. Let's start off with Kubecan. You Alain, I know that you were there. How did it go?
- [03:46]
Alain Renier
Yes, I just switched From Kubecon to Hunt 2 next right over.
- [03:51]
Camilla Martins
It was the week before, so.
- [03:53]
Alain Renier
Exactly. No, it was great. A lot of interesting things to find out. One of the big news for me was the Neoniforce project. I don't know if you've heard about it, they mentioned it during the keynote. It's an effort to build some sort of sovereign cloud in Europe and this is a very major topic in Europe right now and I've been involved on the French side of things with what's called seclude Cloud there. And this new project that we had heard about before is actually backed by the Linux foundation and as you can imagine this is a big thing. A lot of other interesting things. So I was in Salt Lake City before that. I was in Paris before that. I've been going to Kubecon many times over the last few years. One of the things that was very interesting to me also was the presence of opentelemetry everywhere. Everywhere. This is really very interesting and good I think for the industry platform engineering was a big thing also. There are still some solution that seems a bit complex to implement within companies but we're getting there. So something also that was very interesting for me in London, especially compared to the other edition was that I attended several very technical talks. That's not always the case. Maybe I miss them on the program but this time actually a couple of sessions I went out of the session thinking wow, I definitely need to watch it again because it was very interesting, very technical, the kind that I like and I did not get everything so well.
- [05:28]
Camilla Martins
That's.
- [05:28]
Alain Renier
That's pretty much it. Otherwise just a regular Kubecon Europe and very similar to what took place in Paris last year or in Amsterdam to the years ago.
- [05:35]
Camilla Martins
Yeah, the Kubecon schedule is so large it's hard to judge which talks will be exactly what you want them to be. And so that's something common that I hear. So a couple of themes from Kubecon. It sounds like observability, especially telemetry, platform engineering, which I've seen here as well. We had a platform engineering meetup which was really, really crowded here on the next show floor. Excellent.
- [05:58]
Alain Renier
One nice thing also that's a personal thinking, but there was more observability than AI, I would say as a. Oh.
- [06:07]
Camilla Martins
I did hear that it was lower. Yes, yes.
- [06:10]
Alain Renier
And that was a definitive plus. Of course we still need AI and we want AI in observability too. But I think it was much more related to what we actually do with our customers.
- [06:20]
Camilla Martins
So a better balance.
- [06:22]
Alain Renier
Yes.
- [06:22]
Camilla Martins
With the new upcoming AI stuff, which a lot of folks aren't really doing yet and the stuff that everyone is actually doing.
- [06:28]
Alain Renier
Doing.
- [06:28]
Camilla Martins
Exactly. Understandable. And Kubecon is of course a huge event for the cloud native community and is used as an example for a lot of different events. And Camilla, would you like to talk about that?
- [06:42]
Yeah. On cncf and we have other events around the world and well, we have a lot of different communities. So for example, we have the Docker community, that docker is right here too. And we have the Docker community, I'm also a docker captain. And we try to help people to understand containers because it's curious, but you have a lot of people that want to go inside, go deeper in the world of Developer and DevOps, DevSecOps, but they do not understand nowadays how to build, how to create images. And I think the most important thing is to understand the basics. And we go to the containers and we have other communities there. We have elastic that it's also here we have the CNCF community. So we had, right now, in the last month we had KCD event in.
- [07:40]
Rio de Janeiro and that's IBN at East Community Days for another acronym.
- [07:44]
Exactly. And we had a bunch of people almost for a community that just started this year and we had only like three months to organize. We had 300 people and everyone was so involved and in a place that only supports like 100 and a half, 150 people. Yeah. And so we, we saw people standing up and watching the, the talks, but it was so amazing because we have a community in Latin America that wants to help, wants to share. And I think in DevOps, in communities like Cloud Native, the most important thing, it's how we can help support, make people migrate to other careers. And I saw a lot of people changing their lives. So these events like Google Cloud Next Kubecon that I was in Amsterdam two years ago and also in Salt Lake last year, they are so important important to create connection, networking and also to understand that it's not about only your journey but we have a bunch of people that it's only in the same journey. So you have a lot of people that can help you. So that's why I believe in the communities and that's why I support Vince. So we have all of these communities also the Hash Corp community that's also here. Yeah all these, these companies, they are not just in the booths but they are trying to help these people to oh go deeper and go. Go first. So yeah this, it's. It's a hard work I think, I think people that go to the Evans. I don't know if everybody understands how hard it is to does. Does it to do all of this but yeah and that's why I, I celebrate this and and I say to everybody that I, I know that work and volunteer for thanks thanks for your work. Thanks for being volunteer because it's really a hard work to do. Even if it's 160 people, 50 people I meet up for 20 people. It's really hard work because we have other things to do. But we are doing this because we believe on this.
- [10:03]
Absolutely. And you run Cabaret's Community Days Brazil and you do events throughout Latin America. That was a lot of events.
- [10:11]
Yeah. I'm an organizer from KCD Rio de Janeiro with other incredible people but I'm also organizing DevOps days Rio de Janeiro and supporting other DevOps days that they are through Brazil And I'm a speaker in a lot of Latin America events not only from KCD CNCF but also from Google. For example, we had right now we just ended the season of the I doubled International Women's Day from Google and we had in Rio and I'm going in the end of this. It's ending the set that this season. So in the end of this this month I'm going to Bogota to talk to amazing women and we are going to start. We are starting right now also the build with AI. So yeah Google is also celebrating and we have a lot of things at Devrel things and stuff to do.
- [11:05]
Yeah let's dive deeper into the content that happens at these kinds of events. What do folks want to learn along. You were at Kubecon eu you talked a little bit about some of the talks that you saw that were really technical. I'd love to hear more about those. And I'd also love to hear about a talk that I hear you gave about the 10 best practices for kubernetes. Could you give us a little sneak peek at that? What are some of your top best practices? And then we'll come back to your favorite talk.
- [11:33]
Alain Renier
Sure, sure. Yeah. So actually it was a lightning talk regarding GKE and best practices. What can I say? For example, when you get started you need to be very careful with your node pool, the way you create them because they can cost you a lot more than they actually need to. For example, you can select specifically a zone or a region where it's going to be cheaper for you. So you have to evaluate the ratio between the latency for your users and the price you're going to pay. Something as you can do is you can purchase committed usage. So that basically you say I'm going to be using for one year, for three years, you're going to get a discount. It's a nice way to save on the cost. You can also use a preemptible VM is not being called spot VMs on GKE where basically the VM can be kicked at some point, but it will be much cheaper for your usage. So if your workload can handle it, it's very nice and nice way to save money. Another aspect that you can also investigate is all the advanced networking features you on JKE and very often people don't know them or don't know them well enough. For example private clusters where you're going to get only private IP for your nodes. Low security risk of someone accessing the nodes from the outside the authorized networks where you specify which IP address can access the Kubernetes API. What else? Also the data plane v2 which is a very nice feature where basically it's based on EBPF and you have advanced features. One that I particularly like is when you create network policies on your cluster, something is being prevented. Well, nice. The network policy is working but you don't know exactly who was making a request and so on. All that you can get the logs with data env2. That's pretty nice. So again knowing in advance all those network features. Another one also which is very nice is the IP alias mode on on GCP which turns into what I think is called native VPC native cluster. Now on GKE where basically one node will get its IP address but also a full subnet of IP addresses that will be attached to the node and those IP addresses will be used for the pods directly so from a routing point of view, from an overhead point of view, this is actually much more efficient. That's the kind of thing that I think you should know when you're using JKEL and maybe a third one which is very rarely used. And I'm very surprised by that. It's the GKE dashboard. The whole thing has been completely revamped. It's very powerful. You can do filtering, you can do a lot of things. And every time I ask people, I only have a few hands at that razor. Oh, I didn't know I could go there. Usually they get the cluster and they will do the cluster. One advice is to really spend half an hour with the GKE dashboard and then you will always get back to them.
- [14:38]
Camilla Martins
Yeah. Kind of boiling it down to three points that really apply anywhere is node management. Figuring out what machines are in your cluster and keeping it to the machines that you need and the sizes that you need. Fascinating area of cost optimization that I love to talk about. I might ask another question about that. And then networking. Always good to understand how your networks work, especially if you're working in the cloud. If you're working between different regions, there can be costs associated with that. If you have to transfer data outside of the cloud, there's always cost to think about. Definitely something that you want to look into and make sure that you understand as you're getting into wherever you're running. Your kubernetes, clusters. And so we've got optimum node management, networking, networking and observability.
- [15:24]
Yes.
- [15:24]
Dashboards. Ways that you can visualize what's going on in your cluster so that you can control those costs. You can't control the costs if you don't know what you're spending on.
- [15:32]
Alain Renier
Yes. And don't just set up observability and forget about it. Spend the required amount of time to learn how to use them and go back regularly to your dashboard so that you don't lose a touch. Because we also see often people that have not been to their observability tools for a while. They just don't know how to use them anymore. And so they stop using it completely.
- [15:54]
Camilla Martins
Yep. And then at some point you're like, huh, what? Didn't we have something that told me what was happening? Where did that go? I've seen that happen before. All right, Camilla, let's talk a little bit about some of the content that you've seen in events lately. What are some of the big trends that you're seeing around Kubernetes and cloud native? What do you think people want to.
- [16:15]
Learn about a lot about observability like Adam said before, things like OpenTelemetry? I'm watching a lot of these things and in Salt Lake I saw a bunch of talks about it and a lot of trends with AI. So yeah, absolutely. AI is a thing. So how we can use AI in monitoring, so how we can apply it to understand when it's a peak in monitoring and understand if this peak is normal and it's going to be normalized or it's a thing that it's out of the monitoring or maybe also in documentation. AI to help to create documentation because we know that it's not everybody, for example me that we do not like to write documentation as well, but we know that it's so important. So we have Gen AI to well create this documentation not only to us to understand what we did before, what we are doing, but to also document our architecture to our team, to other teams in across work and between other in our company. And also I'm watching some companies that are trying to put AI as a thing to hey, we use AI too. So come on, it's really a thing. And I saw a report that is really interesting from Docker because it's also a thing that we are talking a lot about AI how well it's going to happen to us with AI. And Docker did a report that is so interesting that Genai is helping us with our work, for example to build CLI commands or commands to create documentation or for create tests. Because in our infrastructure environments we know that there's not every place that we create tests for our infrastructure behavior. And we know that it's a thing that we need and also to create code and what kind of code we are creating. So all these models that also we are seeing right now with Genai and 2.5 and others, I think these models that are being best improve it. We can create code that we can go further with our infrastructure as code so we can create best terraform Docker and implement it in our cloud. So I know that sometimes when you get and also to troubleshooting because sometimes when we get a thing that's going wrong, we just got there and we just search it in Google and now Google is better because they have Gemini.
- [18:59]
Also figured it out.
- [19:00]
They have Gemini too. And right now we have. It was a problem because sometimes that error that we had nobody had before. So it was like okay, so what we can do and Gemini it's improving us to okay, what kind of things we can do because we are mixing the models with the documentation that this platform that we are trying to troubleshoot has. So yeah, it's improving us to go best with our applications and to maintain our infrastructure. So I think the all the companies in a way they're trying to put AI but to improve what they have in a way about observability or to create infrastructure or to maintain this infrastructure and well, but the things that I see the most, like Alan said, it's more about observability and how to work with this observability. And like I said a lot of open telemetry and I think that I really love that it's ebpf. I love ebpf. I just met Liz Rice. Liz Rice, if you are hearing this, I love you. We love you Liz. We love you, Liz. So and also about these advanced networks, network policies, how to deal with different protocols and how to work inside the kubernetes with high performance cluster mesh, service mesh, improve the service mesh. Because we are talking about service mesh for a long time time but how to improve the service mesh. So this job with ebpf, they are going really deep dive with that too. And it's amazing. Not only is of island, but other companies too. It's so nice.
- [20:46]
Alain Renier
And just to get back on what you said, I think it's important also to realize that the results we get from AI tools have considerably improved over the last couple of years.
- [20:55]
Camilla Martins
Yeah. Especially in the last couple months.
- [20:58]
Alain Renier
You're right.
- [20:59]
Camilla Martins
Amazing.
- [20:59]
Alain Renier
And we're getting to a point where it's actually becoming really useful.
- [21:03]
Camilla Martins
Yeah.
- [21:04]
Alain Renier
And that's something to.
- [21:05]
Camilla Martins
I kind of love the combination in what you were saying of observability and AI. I think there's something really important to point out there. You were mentioning documentation which I love the call out for AI can be such a great tool to get you past that blank page if you need to write some documentation and you're like, oh, I don't even know where to begin or if you're writing a blog post or whatever you may be documenting. AI can be really useful for just getting you started. And with AI and the ability to generate these things, I think the volume of everything we're seeing is going to increase dramatically. And being able to understand what we have when we're operating in that kind of scale and when there's that kind of scale of things to consume, observability becomes even more important. Again, even though we're talking about observability of systems, I think kind of it all works together, AI is going to kind of explode the amount of stuff that we're looking at in all kinds of different areas.
- [21:59]
So, yeah, I think AI doesn't need to be the whole thing, but a way to use Target. So for example, I don't know how to start the architecture, this documentation, but how can I start it? So it can be useful to create ideas to use. So it's really useful for it. Yeah, yeah, for sure.
- [22:17]
I like to remind folks that content that exists is better than content that doesn't.
- [22:21]
Alain Renier
Yeah, absolutely.
- [22:23]
Camilla Martins
AI is so helpful for that. But you do always have to check it. It can make things up. So you don't want to just AI generate all of your stuff. You want to at least check that it's right. But it can get you far fast.
- [22:34]
Alain Renier
And you want to execute on production what the AI gave you before reading it and making sure.
- [22:42]
Camilla Martins
There'S definitely no security problems in this code. So, coming back to Alain, let's talk about some of those talks that you've seen recently. What are some of the coolest things that you've learned in talks? And we'll be quick with this one. Yeah, well, whatever you can get off the top of your head.
- [23:00]
Alain Renier
So this one, that was a subject that I wasn't very familiar with, which was the csi, the storage interface. I attended a talk that was talking about how to extend the CSI API, but they gave a very nice overview and deep dive of how it works and why this was particularly interesting for me because again, this is a subject I did not really dig before.
- [23:25]
Camilla Martins
And was that at kubecon, the container security interface?
- [23:29]
Alain Renier
No. Yeah, the. No. Container storage interface.
- [23:31]
Camilla Martins
Storage interface.
- [23:33]
What is this?
- [23:35]
Alain Renier
And I was surprised because I went to the talk just to learn a few things and it was much more insightful than I expected. So.
- [23:43]
Camilla Martins
And that kind of. I see a tie in today. It's just what's top of mind, isn't it? The prevalence of stateful applications on Kubernetes is just skyrocketing. Another reason that you need to observe it more closely. And also that storage is becoming really important. I don't know if they made that tie in in the talk, but not in that talk.
- [24:01]
Alain Renier
But that's an interesting point. I also got to see on a developer lounge area talk about. I forgot exactly the name of the project, but it basically is PostgreSQL running directly on the cluster, not even as a stateful set, but with redundancy, with a failover directly. And the demo they made was very impressive. So I definitely need to dig in that again.
- [24:26]
Camilla Martins
While we're talking about storage and stateful, I'll take the opportunity to go off on a slight tangent and tell everyone something that I think folks should know if you've ever looked into. Agonis is a open source project for running game servers on Kubernetes. When I started looking into it with Mark Mandel, who is one of the folks who like really the person who kind of created it, what I noticed is that it's just a really good tool and system for running stateful workloads on Kubernetes. So if you haven't checked that out for your stateful workloads, even though they're not game servers, you might check out the kind of structures and processes that Agona has put in place. It could be useful for all kinds of stateful workloads. So the CSI talk. Any others come to mind.
- [25:09]
Alain Renier
What else? Well, there were several on AI at different level, including here at Next AI for developers because we do development as I mentioned, and starting from asking ChatGPT a couple of years ago, how would I do that? And just finding which framework, which API and getting some pretty good results. Now we see new mechanism, a new application like Windsurf, Cursor and so on, and integration in your Intellij IDE of something like Gemini. Those are very powerful tools and the demo that I've seen were very interesting. That's something also that I definitely want to for me, going to a conference like that, it's a good opportunity to learn about new things and then you want to dig into them as soon as you're back. And the key aspect of all that is are you going to be able to dig into them over the last couple of weeks after the show or months later?
- [26:13]
Camilla Martins
Yeah, and I like it's always good to think about the perspective that you're looking at AI tools from because it's like there are so many AI tools for the developer workflow that I think folks are really interested how they can make their own work as developers much faster. And then there's of course all the AI tools that you build into applications for your users. So there's a couple different perspectives to consider there. So we're talking about like developer tools that are integrating AI. Excellent. And Camilla, you've had some time to think about your top talks and topics recently. What are some things, some of the best talks that you've seen recently?
- [26:51]
I saw some talks are really interesting. For example, I saw on how AI helped on McDonald's in first oh yeah.
- [26:59]
I love user stories.
- [27:01]
We have data governance. It was really nice how to deal with. Because I was talking about it before. It's hard to deal with events and it's hard to deal with a huge restaurant worldwide and they are using AI with that. And it's so interesting to see how AI is helping with different contexts. So I saw that for McDonald's and I saw that for Target too to deal in the context of McDonald's with the. With the food and Target with electronics. And I saw also a thing that it's more academical that it's G co. I think this is the name that it was for images models to images. So how we are dealing with understand if it's a thing that we already have but they are improving it how to understand if it's a dog or not different dogs but they are making it in an academical way. So it was really nice. We had GDE content too. So it was really amazing. And for me the keynote that we had was. It was so nice the VO2 and also the AI agents because I'm more in Google Cloud stuff but because we have a bunch of AI things so we need to learn this thing. So AI agents for me it was thing. So in. In the first day of the keynote one thing that really impressed me was in the moment of. I don't remember but in E commerce that you can talk with the chat and. And collect everything and make discounts and go to the checkout and make the whole process just talking with the. The website. It was a thing.
- [28:51]
So yeah the ways we interact with.
- [28:53]
Computers are changing and yeah absolutely. And how they are now understanding us. Before it was like a thing we talk and sometimes they do not understood before and now we talk and they understand like oh amazing. I'm talking like it's really. It's really a person. It was crazy. And so we had a lot of new ideas and features in developer and the general keynote it was a thing but the AI agents and this news about the VO2 that it's a thing that I'm following since the last year in the Google IO It's a thing that really impressed me. Yeah for sure. And how they are helping other companies, banks and clouds, multicluster and multi cloud content. It's really a thing for sure Some things just Gemini does.
- [29:54]
Yeah I love keynotes at any event Kubecon next whatever conference you may be going to because they always try to have the biggest themes really well represented there. And so you get a lot out of going to one session, even though it's a very long session. I also love that you mentioned the developer keynote because I was actually backstage during the developer keynote helping to run some of the backups. Fun fact, one of the two failures in the developer keynote was real.
- [30:23]
Mofean
Wow.
- [30:24]
Camilla Martins
So go check that out.
- [30:25]
Alain Renier
Okay.
- [30:27]
Camilla Martins
See if you can tell which one was the real failure. It was so smooth, it's hard to tell. So thank you so much both of you for being on today and I'm so glad that we got to talk together at Google. Glad next and make a video. Thank you so much.
- [30:43]
Thank you.
- [30:44]
Alain Renier
Thank you.
- [30:47]
Kaslyn Fields
That brings us to the end of another episode. If you enjoyed the show, please help us spread the word and tell a friend. If you have any feedback for us, you can find us on social media at Kubernetes Pod or reach us by email@kubernetes podcastgoogle.com you can also check out the website@kubernetes podcast.com where you'll find transcripts, show notes, and links. To subscribe, please consider rating us in your podcast player so we can help more people find and enjoy the show. Thanks for listening and we'll see you next time.