 Let's get started. Welcome to Observability Day. And that's to my wonderful and esteemed French colleagues, that is the last time I will butcher your language. So let's get started by thanking a few of our sponsors, whose support is critical to put on events like this, which is now the fourth observability day, I believe. New Relic, honeycomb.io, OpenSearch, and Grafana Labs, also sponsoring. You can, yeah, yeah, no, give yourself a hand. You can check them out outside. Couple quick words. Observability Day is really meant to kind of bring together all of these observability projects under the CNCF banner and let us share our successes and talk to each other about where we've been able to push forward in getting better observability, right? It's not a destination. It's a journey, as we like to say. The projects that you'll see represented here is everything from Prometheus, the gold standard for metrics that's been around for so long, Jaeger, Fluent-D. Also, there's a lot about newer projects, such as open telemetry, Thanos, and Cortex. This is the fourth iteration, as I said, of Observability Day. This started as Open Observability Day in 2022 in Detroit. Last year in Amsterdam, this was one of the best attended Colo events. We had another one in Chicago that was literally so full, they were turning people away at the door. So this year in Paris, we're actually extremely fortunate to have not only some very large rooms, as you can see, but also two of them. And they're unfortunately on different floors. So I hope everyone brought their FitBit because you'll be getting your steps in going between sessions. This is really a collaborative experience for all of us, not just speakers and presenters, not just chairs and program committee, but also all of you, people that are attending, people that work in this space. So when there's breaks, say hi, introduce yourself to other people. If you have a question for a speaker, feel free. I can say, as a speaker, the thing that I want to do the most is I want to talk to people. We wouldn't be up here otherwise, so please, if you have a question, go up to a speaker, ask them, we're all very friendly, we don't bite, I promise. And remember, we are all learning. There is never a point where your learning journey stops, so you're gonna keep on going. Quick note for speakers, there are the mics up here. Also, if you'd like a lapel mic, you can find the AV people at the back of the room here or upstairs, and they will get you sorted. We have a ton of talks. I don't have time to go through them all in depth because we have quite a few project updates to get to, but please keep in mind when you're looking at the schedule, make sure you're in the right room because you're either here or you're up on 7.3. So you go to the left as you get off the escalators and it's right to your left, actually, when you go back into the room there. Finally, there's a code of conduct. You can scan this QR code to access and review the CNCF code of conduct. Generally, just remember the golden rule, treat people as you would like to be treated and treat everyone with kindness and respect regardless of their background or anything else. Captioning and translation services are available via Wordly on your personal device. You can scan this QR code, choose your language. Session ID will populate and it's like a tend. Captioning begins when the session does. We will have refreshments, snacks, lunch available throughout the day in front of the B and C rooms and then in front of the ENS rooms on level 7.3. There will also be a reception tonight from 1730 to 1900 hours. Love to see you there. During lunch, we'll have some table topic signs so you can write, hey, you wanna talk about EBPF or you wanna talk about profiling, then you can write a card, put it on your table, and have people come and chat together. It's a fun way to make connections. Couple final notes here. There's a social media hashtag of Oli Day, virtual discussions in tag of observability on the CNCF Slack. If you're a speaker, please try to be at AV five minutes before this are your presentation. And at the back of each room, there's some stickers and other fun stuff. All right, talked about that. Thank you to our program committee. Give them a round of applause. They are, they put a lot of work in. This is, this has some of the most talks submitted for any Callow Day, for any day zero event. It's a really challenging job every year to go through and pick out the best talks for everyone. So with that, let's move into our project updates. So first off, we have Brian from Prometheus. Hello, hello, thank you all for coming. So yeah, this is me, I work for Grafana Labs and a quick update about Prometheus. Well, let me see, who uses Prometheus? Who's never heard of Prometheus? Nobody, okay, so I'll skip the next four slides. Prometheus is 10 years old. There's a documentary about it. And we're still going strong though. We did eight releases in the last 12 months. So pretty steady cadence every six weeks we do a release. There's also the long-term support version of Prometheus, if you're not aware of that. So if you don't wanna take the absolute latest, we're updating that. Prometheus is not just the monitoring program in the center, it's the whole ecosystem of alert manager and exporters and clients and so forth. So we did a 1.0 release of the Java client. That's pretty much a rearchitecture of that. Check it out if you are interested. What else? I wanna mention native histograms. We've been talking about this for a while, but they're nearly done. So the point being traditional Prometheus histogram is very low resolution. It's kind of the blocky thing on the left. And what we're calling native histograms are nearly infinite resolution, much more fine grained and more efficient. So those are still declared experimental, but pretty much ready to go. So definitely take a look at that. Prometheus got smaller. I did a whole talk about this at the last observability day, so I just wanna, again, if you've not been following, it is pretty much half the size it was two years ago. And one more slide. We are preparing a Prometheus 3.0, which is scheduled for PromCon, the Prometheus conference, September time. These are, Open Telemetry is a big theme of Prometheus 3.0, and I won't go into all the detail. We are holding a maintainer session in the main conference on Thursday, Thursday afternoon. We have a booth in the main conference starting tomorrow. So if you wanna talk about any of this detail, come along or check us out online. That's my update, thank you. And next we have Eduardo Silva from the Chronosphere these days. Yeah, it's Chronosphere now. Talk about fluid bed. Thank you. Hello. Good morning, everyone. Happy to see you here, and I'm glad to see how this conference is growing. As Austin was saying, we used to have just one room, now we have two. Maybe next year we have three or four. We're full pack. So let's talk about the news of the project. I know that we have some users from FluentD, from Fluent Bed, so and there's always some a little confusion about, hey, what is the project and how this works? Actually Fluent Bed is a CNCF graduated project under the FluentD umbrella. And as part of the problems that Fluent solved is everybody, what you face in observability or where you are trying to deploy applications. Deploy is like one of the steps and the next one is to monitor what's going on in your environment. So how do you solve moving data from sources to destinations? And that is with FluentD and Fluent Bed does. It sits in the middle as a way to connect different systems, different telemetry data and also solve this problem, which is not just connectivity. Also buffering, retries, scheduling, moving data at scale, it's a bit of challenge. And the way that this works is like you sit in the agent or any type of product in your environment and then you connect to your own destination, your own vendors. This is fully, fully agnostic. But also when things start to run at scale, you need to have a way to deploy this in a very smart way where you might prioritize different things like, for example, do not lose data. I'm okay to lose data, but I need to move the data faster or I need to process the data before sending the data to the final destination, right? And the agent, Fluent Bed and FluentD, allows you to send this data to one or multiple places. Also, this is a common pattern architecture when you deploy Fluent or any type of tool that allows you to concentrate data. This is what you're seeing in the graphic right now in our world used to be called an aggregator. It aggregates data from multiple endpoints before sending the data to the final destination. And one of the reasons for that to exist is because sometimes you want to control the data flow. Maybe if you have a thousand machines, you don't want the thousand machines sending data directly to your backend because it can blow up and maybe you need something in the middle that control, validate the data schemas and so on. And one of the values of Fluent is that also it has been extended not just to work with Fluent. It works with Prometheus by doing scraping with open telemetry, locks, metrics and traces. So it becomes a real neutral way to connect different systems, right? Where you go in production to any type of company today, you will find that you have data from SysLock, from ProLock, from Prometheus, from MySQL and everything is a different format, different transport mechanism and Fluent allows you to integrate those and solve your problem. As of today, Fluent bit has been downloaded 13 billion times, deployed on Kubernetes Cluster. This is insane. If we do a checkpoint from one year ago, we are two X now. So that means more users, more bugs, more futures, more improvements. So it's good to have this pressure to get this adoption at a production grade where now we need to go for the next big steps. And one of the next steps is that this week we are releasing Fluent Bit version three. And I'm happy to hear that also Prometheus is hitting version three. We should do a joint party from the same ecosystem. And one of the biggest thing of B3 and maybe it's not a thing that the user will face, oh, now I'm using HTTP2. It doesn't matter, right? So you just want to deploy your application, making sure that you can connect the stuff. But HTTP2 is a big deal to integrate system where if you are instrumenting your applications, for example, with open telemetry and you want to rely on HTTP2, now Fluent Bit will be able to collect data by using that type of transport. Also we are doing a lot of internal improvements for performance mostly and how to facilitate the way that people can extend this agent when you are modifying the data, right? I don't know, we're not going to talk about internals but basically when you receive data you need to encode it, decode it, serialize it and convert from different formats. The most you will improve on that area, the best performance you get and performance is not just a speed, it's also consume less CPU and less memory. We have achieved a new processor. We have filters and processors. Processor, we achieved those interfaces like a year ago which allows you to modify in a more easy way the content of log and traces. Before you have different plugins as filters in Fluent Bit to modify the content, but now we're achieving a very easy way to just modify, insert, rename keys of logs and traces. And with different type of operations, even if you want to extract the content from a specific value with a regular expression, you can extract that and convert it to a key value pair, basically. If you are playing with metrics, sometimes we get users that say, hey, I don't want to hold the metrics that is being collected. I would like to just exclude or include certain ones based on criterias or prefix by names or regular expressions. So now we have a new processor for that specific thing. And now we have chipping a new SQL processor. You will say, hey, SQL, are you running a database? Ready? No, no, sorry, SQL Lite or something in Fluent Bit? No. Basically, our SQL processor allows you to select specific content from a log by using an expression with SQL query language. So you can select keys, rename them. You can put expression. You can create conditionals, for example, for values, for strings. And in the next couple of iterations, we are going to ship aggregation functions. Aggregation means if you want to create a window of time, and in that window of time, you want to count what is the maximum value, the minimum value, and just start collecting the data that you really care about. The first problem that we solved at the beginning was connecting A to B. Then we added an X problem, which was processing. And now we are doing a high level where we are putting more logic for the user so you can have a smooth experience. And also, when talking to our open telemetry users, one of the biggest challenges that we found is for somebody who's generating data from an NCK in open telemetry and sending data in open telemetry format, it works fine. But most of our users, mostly coming from logs, they said, hey, I'm not instrumenting my applications, but I'm shipping data in different formats. How do I connect to open telemetry? And for them, we need to ship some solutions. So now our new open telemetry output connector has been improved in order to support more processing capabilities to adapt your own logs to the expected format by open telemetry. And as a final message, we always do a big release of flu embed, which is just also a new T-shirt. So on this T-shirt, we try to give it for free at KubeCon. So please, we will be around observability day. We have like 50 today, so make sure that you find us outside, and you grab your T-shirt, or also in KubeCon. So thank you so much. And I'm going to. That's the lesson. Thank you. Thank you. All right, so let's get through some open telemetry updates real quick. We're going to be pretty brief. We have a maintainers' track talk on Wednesday that will go into more depth on a lot of this stuff, but really the three big new things in open telemetry we want to talk about right now are continuous profiling events, real user management, and then a lot of stability around core signals. So we recently merged Oteb 239, which brings in continuous profiles as a new signal to open telemetry. We expect an implementation this year. There's some work happening with Elastic contributing some code there, which will dramatically accelerate it. We are also working really hard on stabilizing an event spec to accelerate several different things, including real user monitoring, so a lot of improvements to the JavaScript libraries. This is going pretty well. We're expecting that around Q3. Finally, last year we announced Logs, Metrics, Traces were stable. Great. So what does that actually mean for you? Well, now it means that there's four languages where that's actually true. So that's been a long journey to get there. But we're also expecting Golang to follow this year, which I know means a lot to the cloud native world. We're also working really hard to stabilize semantic conventions. I know a lot of people have dangling dependencies on those. Again, if you'd like to talk more about any of this, we will have a maintainer's track talk. And one more thing, we're applying for graduation. So if you're not familiar with the CNCF maturity levels, you have sandbox and committing graduation. We believe that the project has matured enough from a governance perspective, and it's in production worldwide. So it's time for us to kind of take that next step and move towards graduation. There will be some additional things about this, which again you can learn more about at our maintainer's track session on Wednesday. And we expect to conclude this by KubeCon North America this fall. So I don't know what the booth number for the observatory is, but it's the big thing that says open telemetry observatory on it on the show floor. And please join our event channel on the CNCF Slack Hotel KubeCon EU24 to get more info. Next up, we have Jonah and Pavel talk about Yeager. Yeah. Hi, everyone. My name is Pavel. I work for Red Hat. Jonah at Ivan. Nice to see everyone. Jump into a couple updates on Yeager. Yeah, so the main update is work around Yeager V2. So we are planning to rebase Yeager components on top of an open telemetry collector, which means that we want to essentially build another distribution of an open telemetry collector with Yeager components. So all the storage layer will be implemented as open telemetry exporter. We will produce a single binary compared to all-in-one query and collector that we have right now. And we would like to, as well, align the configuration for that with the collector so there won't be flags. It will be only the configuration file as we have it in the hotel collector. And we still want to be compatible with the existing storage layer, so we will be able to migrate to V1 without doing any migration on your database. Yeah. Awesome. Thanks for those updates. And so a few other kind of new features that we've announced since the last KubeCon. We fully support Elastic 8. That's something a lot of folks were asking for. We've also built a whole bunch of new capabilities in the UI. So if you haven't updated your Yeager, definitely do so. It's a pretty easy update, and you start to get some of these new capabilities. And then finally, on the roadmap side, just kind of further taking what Pavel gave you an update on, we're going to have a beta soon of V2. We're not on V3 like some of the other projects, but we've been very stable on our V1 for quite some time at Yeager. The other thing that we're going to be releasing is native support for Clickhouse. Today it's a partially supported implementation, but a lot of folks using it. So come to either the booth. We're there in the afternoons. You'll see a Yeager booth. Or you can come to our maintainer session on Wednesday. I believe it's 12.10. So please stop by, and we'll get into much more detail than this quick update. And thanks very much. And there is, as well, Yeager Country first on Friday. If you would like to help out with the project. So hi. Hello, everyone. I'm Sasweta Mukherjee. I work at Red Hat. I work on monitoring platforms, largely based around Thanos. I'm a Thanos maintainer as well. And you can find me at the Sasweta M code pretty much anywhere. So the Thanos community this past year was super active. We had a bunch of new people, new contributors, new LFX mentees, even new maintainers. And we are very excited to share some of these major updates with you. So starting with query. So this past year, there was a lot of focus on Thanos query. So we released a version with distributed execution that you can enable with a query.mode flag. And essentially what this distributed execution does is that it has functionality to split your queries and delegate them to either your leaf queries or even child queries. And as PromQL has mostly aggregation operators, you can see how it would be largely beneficial because each query has to only process very little series as it goes up to the root. We also, for a new engine, we had a lot of progress. It now supports up to 95% of all PromQL expressions. So the Volcan engine has become a really nice and viable alternative to the Prometheus engine in the context of Thanos. And we also introduced a native query tenancy model. So with HTTP headers, you can request data for a particular tenant. And this can be paired with any OSS authentication authorization mechanism to give you a very complete authentication plus tenancy model on the Thanos query, on the read path, basically. We also introduced a few changes in receive. A major one is availability zone-aware replication. So what this does is that it assigns every single Thanos node in a hash ring to a particular AZ and then data can get spread evenly across your region. And this just boosts your ingestion path as low as your disaster recovery scenarios and so on and so forth. And there's a lot more. There's selective index caching for storage. There's TTL. There's lazy downloaded index header for storage. There's extended functions on the query and the query front and even reloading using signals from the sidecar. And the Thanos community has been just super active. We've also had a lot of mentees that we've built projects with like dynamic switching of engines on the query and explanation and analysis and so on. And finally, for the first time ever, we actually are hosting ThanosCon at the CNSCubeCon Paris. It's on 11.7.3, Room E02, where we get into depth with a lot of these details. The Thanos community shares what they love about Thanos and how they're using it to build cool stuff. So definitely feel free to visit. We also have a kiosk, BP18A, I think, on the CNCF project pavilion. So if you have ideas about Thanos, if you're excited about its future or if you just want to get involved, feel free to reach out. Thank you. Thank you. Is Friedrich? Yeah. OK. So I'm Friedrich Gonzalez. I'm a software engineer at Adobe. I'm going to give some Cortex updates. I could go around down the Prometheus updates again because pretty much everything that has happened for Prometheus has happened for us since Cortex is a project that uses Prometheus in its code base. But anyway, so we have not been lazy. We have not just been using the good features from Prometheus and Thanos, but we also been adding some great new features in the last year. And here we have some of them. I'm not going to go through all of them because of the time. Interesting is query priority, right? So we have a way to specify specific queries which give higher priority. And well, that's a lot of stuff there. We also have in the roadmap for this year a lot of new features, OTLP ingestion. We have been working on that for a while. I myself have been working on that. And also the native histogram support as well. We have a demo for those two features on our talk, which is the next slide. We have three things that are important. You can come to the talk that we have tomorrow. And learn more about Cortex and all the users, what the users enjoy with Cortex. And also we're going to do an introduction there. If not, you can come to the booth. We have a booth. You can come talk to us in the booth. These are the times that you can go there. And if you're not in the conference right now, you can go to the CNCF channel and you can ask questions there. That's where the users are. All right, thank you so much. And that is that. So thank you all for attending. Please check your schedule for which session you want to be in and enjoy Observability Day. We'll see you back here at the closing.