Peter
00:00:00.000
I think it's very important as a founder to think, what you're really best. at Right. And there are some people who are, I don't know let's say, born salesmen they're very, very kind of, quick to monetize any hype out there, right? They would be on it the next day. And if you think about, you know, Nikolai and frankly me, we are not that kind of people, Hey, you know what, being a hype expert is not us, But we can build you a generally good. Observability product. And that is what we're going to do.
Darin
00:01:36.996
Viktor. I'm gonna give you a handful of names altogether, and then I want you to tell me after I give you the names, what comes to your mind first. Are you ready?
Darin
00:01:48.636
Datadog. Dynatrace? No, no, no. Hang on a second. Datadog, Dynatrace, new Relic, and Gana. What are those four things? Okay. OpenTelemetry. That's one of the things. Uh, gimme something else.
Viktor
00:02:16.071
Excluding Grafana. So kind of if you exclude Grafana from that list, then it's me go going bankrupt. Yes.
Viktor
00:02:27.621
I dunno. I know you're leading me towards some answer, but I dunno what the answer is, man.
Darin
00:02:32.856
They're all cloud-based. all cloud-based. Graf. You have an option I, I will give you that. But the other three, the first three for sure. are all cloud-based. So on today's show we have Peter Zeon from Coroot. Peter, did I get your name right? Even close. Okay, good. Uh, he's a co-founder at Coroot. Coroot has a solution that is in that same family. Peter, correct me if I'm wrong,
Darin
00:03:00.021
However, unlike the first three, we'll still leave Gravano out of it. we can run it locally, right?
Darin
00:03:12.891
Well, okay. So it's possible you could go bankrupt. Lemme, I have, so let's pause it there for a second because again, we like poking at the big three. My question is, couldn't we still go bankrupt buying petabytes of disc and the power that cost to run them?
Peter
00:03:30.955
Of course, in, uh, this regard. Yeah. But that's, much harder, to do that if you are actually just paying on the, infrastructure costs, especially if those infrastructures run efficiently compared to, having to pay those, uh, many times, uh, markup, which the cloud vendors especially have on the infrastructure cost to keep their margins.
Darin
00:03:56.428
So, okay. I went down an earlier or a question I was gonna use later to now, so let's pull it forward. Viktor, you wanna say something? Go ahead.
Viktor
00:04:04.843
Yeah, I could now turn back your question back to you kind of. I could also come off with the word rich, rich, rich. And then when we jump into, into Peter, then ka poor,
Darin
00:04:22.318
okay, fair enough. We'll, we'll let him answer that later, but, so Coroot has, can we call it an open core model? You have an open source community version, and you also have the enterprise or paid versions and blah, blah, blah.
Darin
00:04:39.598
Okay. I'm assuming let's, let's talk about Coroot itself, but then I wanna talk a little bit about the business as well. 'cause, you know, in this time of, I can spend a week used to, we would joke around, it's like, oh, we'll go to a hackathon and build something this weekend. In today's world, you can almost actually do that. why did Coroot even come into existence with. Uh, okay. Let's compare it to Grafana, right? 'cause that's everybody else we have to sort of set off the block. Why was it important for Coroot to be able to run self-hosted.
Peter
00:05:09.742
Well, I think, a lot of that comes from, uh, differentiation. as you mentioned, there are a lot of, clouds, products in, observability space, right? So, so that space is actually, extremely crowded and it's, uh, hard to, get, into that space where if you look at the, observability for self-hosted solution. that is, uh, different market. It's not, so crowded and I think that is also the space where the open source, solution has, much more benefits. I would say so at some po at some point, we may also offer the cloud-based version of, Coroot for Deployment for those people who are. Looking for, simplicity. but at this point we decided to only provide, as you mentioned, the open source, version, our community edition, which is a, we are, uh, trying to be, uh, sort of like the, the best of breed when it comes to open source observability solutions, right? And then, the, enterprise version, which has some, uh, additional bells and whistles, which we hope enterprises would choose to pay for.
Darin
00:06:15.825
So, lemme back out a little bit. Can you compare yourself to the open source version of Grafana? Like, I think Grafana, I think only dashboards,
Peter
00:06:26.940
so, uh, first I would say what, uh, Grafana, and Coroot are not direct competitors. We have number of people who would use, Coroot and then they would use Grafana additionally to, uh, get the data from Coroot and visualize, right. Get more. custom visualization, and so on and so forth, right? Some advanced, uh, people, do it right. If you think about the Coroot, right, and, oh, sorry. If you think about Grafana and actually what got excited me years ago when it just. Came to market is how fantastic, sort of like a Lego block it is, right? It's really allows you to build your own visualization. Hey, you can get a data from about anywhere and visualize it in so many different ways and with plugin even more different ways, But that also is exactly that Lego experience. You can build almost anything, with Grafana, but you need to build it. now, if you look, at, this case Coroot, is a solution not for somebody who wants to spend their days building observability system. But it is focused about a developer who has an application to run. And in the modern world, You can't really have a successful application operations if you don't have observability. Things are just very complicated. You know, think about the microservices, you think about talking to external APIs and so and so on and so forth, right? So if you want to get that observability, but you want to get that easy, that is very correct. What comes in. Coroot is also what you can think about the opinionated observability, We're not just saying, Hey, we are going to show you whatever you want to do. Uh, right? We are saying, Hey, if you want to operate your system, effectively, if you want to resolve a problems quickly, you need to look at those things, This is a methodology you follow, And this will give you results, uh, simple and quickly as possible. For certain people it is awesome, For some durability X, we can say, oh my gosh, I don't want you to tell me how to look at my data and how to use it. Well, in this case, they can either build their own systems, you know, altogether, right? Or, complement Coroot if, uh, uh, something like Grafana.
Viktor
00:08:49.689
I think Grafana did itself miss favor that they named the company based on the first product that they worked on, right? In reality, Grafana is realistically Grafana, Prometheus. And, uh, you know, whatever they call the tracing solution and so on and so forth. Right. Lucky
Peter
00:09:09.674
That's right. That's right. But that is even more building blocks, right? You have visualization building blocks. You have visualization block for time series. You have a visualization, block for, uh, for tracing and so on and so forth, right? And then you go and, put that together and you can do absolutely awesome things with it.
Viktor
00:09:28.991
Yeah, so this kind of like, uh, out, if I would, if I understood correctly, car, the difference is that Coroot is the out of the box solution in a way. while Grafana. Family, let's say, would be, hey, you install those, uh, five things and, uh, you spend, uh, you need to spend a few hours looking for existing dashboards that you want to use, and then potentially go and modify them to fit your needs. Right? This is kind of, out the box, right? Open the box, and you have observa abilities. Is that more or less the direction
Peter
00:10:03.701
yes, that's right. Right. But I think it's also important, like a philosophical difference, right? Because if you look at a lot of, companies in the vulnerability space and not just Grafana, we have this philosophy as more is more. If you get, you know, more metrics, more logs, more tracing, right? And we give you more dashboards, Hey, you know what, like a thousands of dashboard is better than, uh, than 10, And yes, for certain kind of user, it is, uh, absolutely, uh, a case. in, uh, our case. We think what the less is. We think about, hey, how to help people to avoid that cognitive overload, right, if you will. How to we give you the minimum essentials, to achieve your goals, it's not only about visualization, if you think about important part of, Coroot, right? And another problem we see in in observability is observability rollout. You mentioned, for example, OpenTelemetry, you know, fantastic, technology, but that is also not an easy to roll out, If you really want to roll it out, you have to, configure your applications to emit OpenTelemetry traces. For some new applications, that's maybe easy. If you have some old applications, we don't have a team maintaining them or some proprietary applications that would be invisible. in Coroot we are using eBPF, so you can various technology which allows us to get visibility for any applications without any additional instrumentation.
Viktor
00:11:30.048
So how does that work? Do you kind of run an agent, let's say micro Kubernetes cluster that automatically instruments every container pod what's not, uh, over there,
Peter
00:11:43.973
That is exactly that? So you would have an, uh, agent, which will, uh, instrument every container, every pod, uh, right. It'll understand how different microservices talk to each other. And so how errors and latency, propagates, If you are running that in VM environment or bare metal, the kind of the same thing applies, right? We need to get an agent, uh, all the node. But this agent, I think what is important, it doesn't require. Configuration, right? It's really, the instruments, things, optim automatically, right? And there's a lot of like a sophisticated, things like, like say, oh, we are looking for example, what the, something which looks like a log file is open, And we'll, Optim automatically detect that, and then they'll start analyzing that as a log file. You don't need to because of that. Understand what kind of applications you have and wherever we have a log files and where those log files, go, if that's files right, and, and so on and so forth, right? So, we are trying to get as many of those kind of, uh, you know, nice little tricks, as possible to really reduce, the things, what, uh, users need to do to get, uh, their observability.
Viktor
00:12:56.406
But then if, if I have, let's say, application that I instrumented because I need traces that are not generic, but you know, uh, I I want to trace this specific function in this specific moment in time, whatever. Right. do you still kind of like, is the less is more mandatory? Or if I give you, if I already have traces, you pick it up still,
Peter
00:13:21.476
Yeah. Oh, uh, yes. Coroot can consume OpenTelemetry traces as well. Right. So if you, yes, because you can get more if you chose spend the time instrument in application, if you have that information, Coroot is able to consume it again. Like also if you, spend time, for example, to get, uh, give Coroot credentials about, from a database, right? It'll able to get more data, which it can get directly from database like MySQL or, or Postgres. Right. If you don't, it still can get a lot of information, uh, about, errors about, performance, right? Based on, on the traffic of how your application, you know, sends queries to a database and gets responders back.
Viktor
00:14:04.783
So it's almost kind of your to observability what Docker was for or is to orchestration in a way, right? Kind of, you know, the major difference between Docker and Kubernetes would be that Docker and Docker's form in the past would be, Hey, this is out of the box, kind of five seconds and five minutes and you're up and running. Or you have other extreme Kubernetes kind of, you get amazing results, but you need to. Roll up your sleeves type of solution, right?
Peter
00:14:34.505
that's right. That, that's right. Right. So our focus, uh, in this case is to make sure you get the easiest solution, uh, out of a box, The other thing, what we, uh, didn't talk about right, is what a lot of people are looking for, right? Because we often talk about logs. Traces metrics, uh, right. Uh, and how that kind of visualized in the dashboard, But, uh, what you think in, uh, in Coroot is really the outcome people want to achieve, right? And that is often, Hey, I have a problem. I want to fix that as easy, as quickly as possible. And that is where, with, Coroot we do a lot of, root cause analysis. where we are able to highlight what exactly went wrong and I think that is the most, important concept here, right? what we saying, Hey, if something goes wrong, we are not expecting you to go through a dashboard, right? We are expecting to tell you what exactly went wrong. And then also, use their, AI LLMs, right? To give you possible fixes and we can do some pretty. Pretty good stuff in this case, which you probably can give you useful advice to cover 80% or more without actual a lot of human analysis.
Darin
00:15:56.202
Now you brought up AI there. We've already talked about being able to self-host everything. Can we self-host the AI as well or am I gonna be stuck? I mean, I, I can, but am I gonna get the results I would get from, I mean, wait a week and you get the next big model. And I don't think the self-hosted models are there yet.
Peter
00:16:13.725
Well, uh, and I think, uh, uh, that is an important, challenge in this case, right? So we are given a lot of flexibility to the users, uh, in, uh, this regard. Uh, right, first of course, you don't have to use, uh. AI functionality, right? If you are saying, Hey, there is no freaking way I'm sending any of my data to ai, right? Okay, cool. You can do that right then. Uh, you can, uh, use locally hosted, models or, uh, you can, uh, use some of, uh, external models. number of people we can see. If it's a larger enterprise, they may, host some of, uh, those models on Amazon Bedrock, for example, which are can give you kind of a good result, but that also can be easier, operate compared to, you know, just rolling out your own stuff on GPUs directly.
Darin
00:17:01.493
I wanna go back to eBPF real quick. Uh, we've talked about eBPF on the show before. Could Coroot have existed without eBPF?
Peter
00:17:11.622
Well, not in this form, because I think that, eBPF, is, awesome technology in terms of what it's, uh, being powerful, but it is also very safe. In terms of how it's engineered, right? Because if you look at what Coroot was doing in the past, let's say before EPF was, existence, you could could say, Hey, we are going to run, uh, like a special, uh. Kernel model, Which is going to have functionality, which is kind of somewhat similar, right? Allows us to intersect everything. But that is a very scary proposition, right? Not a lot of, uh, people say, Hey, we want to inject some kernel level, stuff in our production boxes. That kind of sounds very dangerous, right? And also considering how many different Linux, uh, uh, versions out there and flavors out there, it's also is very hard to maintain and keep safe. that is why you can see a lot of companies which would, do some sort of, kernel level work. Not only, in terms of, observability, but also for example, security or network filtering. We are saying, Hey, you know what, instead of us running kernel models, we are trying to implement as much stuff on eBPF as possible because it's very high performance, but that also is, uh, uh, very safe compared to alternatives.
Darin
00:18:30.399
So do you just produce the eBPF files? I can't remember what they are. It's been so long since we've talked about 'em. Do you just produce 'em? And so when I do the install of the quote unquote agent, I'm assuming there's some sort of agent, is that correct? Okay. So that, so when the agent installs, then it drops in the eBPF files where they need to be, and life goes on.
Peter
00:18:51.409
Yeah. not exactly. So let's kind of, drill a a little bit into how things work, First, We talk about eBPF, but eBPF is not the only technology which, uh, Coroot uses, right? It's use a lot of standard, technology as well, and it's kind of connects that together with what eBPF is, right? Like for example, eBPF can intersect the. Traffic going on, right between two connections. But then you can use, what is exposed to Linux Pro Fast or wherever to really, understand what process it is, right, to maybe to what container that belongs and, you know, and, uh, so on and so forth, right? So you have a lot of information, sources out there, which are kind of being, glued together. But eBPF is a core of it. Okay, now let's look at what is eBPF and how, that, operates. The functionality we use in eBPF is specifically is able to hook up to any function calls inside the kernel or inside application, right? So, for example, we can hook up to where the connection is initiated and to say, oh, that is connected to that host. And it took, you know, so much, so much time, right? In this case stuff, uh, stuff like that. Now, eBPF as a technology, it actually consists of two parts. There is a kernel level part, and there is a user level, part, So in a kernel level, that is a very small function which generates events. So, or, or counts some sort of stats which are stored in that kind of fix it memory in the kernel. So there is no kind of files because files are. Slow. Right? And files can potentially, block things. Uh, if you are trying to write a storage and storage slow, right? And so on and so forth. It's all very fast on in memory. And then there is an agent which asynchronously can consume that data. So for example, kernel has, um, ring buffer, right? And then the client would consume events from this, the client in this case being an agent, consume this data from a ring buffer and then sends that to the Coroot server, So that's how pretty much operates.
Darin
00:21:01.597
so we have our data flowing from our Node.js. We'll call 'em that because they could be a physical server, they could be a pod, whatever it is. I mean, a pod's on a physical server. But go with me here for a minute. We have all that data flowing into the covert server, and I'm assuming that's where the magic pixie dust is applied.
Darin
00:21:23.407
And then I, I can choose to use Grafana if I want to for visualizations, or I'm also assuming there's some basic visualization. I'm, I don't mean to disparage it, but it basic comparative gr what Grafana can do from a dashboard. Is that, am I
Peter
00:21:38.362
Yeah. Well, I would say it's, um, Not just basic, but it is different, right? Because if you look in a Grafana approach, right, there is a lot of that is based on like a fixed dashboard, If I want to have a graph which looks and kind of in particular way, it works, uh, pretty well in case of Coroot, there is a lot of those dynamic visualization. One of the core things, for example, we can visualize your map, is about how different services are talking to each other, how traffic between them, uh, sort of like flows in real time and so on and so forth, right? And that is like a pretty, unique, uh, visualization, which would not be easy to repeat in Grafana, especially without, uh, without plugin, But now if you look in this case and say, well, you know, I want to, plot some. Specific disc utilization metrics, right. Or, you know, how different host group CP usage looks like. Right. kind of sum to group together, right. Average some way. Uh, graph, uh, uh, where Carro has some custom dashboard functionality, it's not nearly as advanced as, and and that is I think, where It is awesome. And then you, you mentioned this kind of like a special, uh, magic, right? The, the important thing. What, Coroot does is it really is very good at, uh, uh, linking things together, how things, propagate where errors originate and where those, errors can impact the upstream services.
Peter
00:23:08.368
That's right. Uh, uh, that's right. Root cause analysis obviously is using that, but you are also able to visualize that in a different way to say, okay, like I am seeing what this service, having an errors, let's, uh, look at that. Oh, uh, the, it's latency increased to that database, right? That to that given microservice, uh, which is slow because a database is being slow. You can look at that, manually, right? Through visualization tools, right? You can do the root cause analysis, which is, uh, which is going to do that, uh, uh, for you.
Darin
00:23:41.417
So back to the visualizations real quick and we'll, we'll definitely move on from that. pick the right tool for the data you want to look at is what I'm hearing. It's like I, if I wanna see the. The time series of CP utilization or disk and and aggregate over servers. Grafana might be the better way to do that. Or if I already have dashboards already liking Grafana, great. I can bring those over ish.
Peter
00:24:05.307
that's right. Right. And I think this is something which is also important and how observability tools are different from some other things, right? Like if you are using database for example, you typically have to make a choice, right? Oh, am I going to use Oracle? Is it going to be Postgres? Is it going to MySQL? Is it going to Mongo? It's not practical to use all of them together. when it comes to observability. It's typically, you can use multiple tools, Especially if you think about there are some observability tools, which are sort of, you know, good visualization. There are some fantastic, observability tools, which are done for, you know, interactive diagnostics, right? Maybe common line, right? And so on, so forth. And you can think, many organizations, if you look at them, they would be using, uh, multiple, observability tools for, uh, the different purposes. Right, so it's not either, or a situation.
Darin
00:25:00.959
If we think back in the times before OpenTelemetry was really adopted by a. What we were calling the big three earlier. it was that way. You pretty much picked a tool or every silo within an organization picked a tool and did their own thing. But once OpenTelemetry came into play, that helped break things down a little bit. And then once eBPF was inserted into everything and people started using it, that also broke down more walls. What's the next big wall that needs to be broken down?
Peter
00:25:36.453
Well, I think you're right in terms of kind ofer, certain walls, right? But I think it all goes into the level how deep you are actually, going to go, right? Because, I have a lot of connection to the database world, And in a database world, there is a lot of stuff which doesn't quite, uh, maps neatly on things like, you know, traces or, uh, or metrics and, uh, so on, uh, and so on and so forth. for example, if you look at the, queries, right? You don't want to trace every single queries because there may be, you know, like a millions of seconds, but you want to somehow cluster queries together, right? So we can analyze how they perform, Their base buy, for example, execution plan, right? Or some, some other things, right? There is some rather complicated things. How we want to, Analyze deadlocks, right? For example, which is kind of really of us kind of fancy, graph theory under it, right? Which has its own, visualization, tools for that, right? So that space has a lot of, uh, additional, uh, uh, special tools, I would, uh, highlight, right? Or another tool in which I. Mentioned, right? Which is yes, based on, uh, OpenTelemetry, but you think about, bpftrace, right? For, uh, kind of standalone analysis, right? Or something like, inspector GA Gadget, which is focused on, Hey, let's see what we can get from a system now, what is going on? Right? Where to OpenTelemetry, right, or, or metrics captured from, eBPF are. typically something which you keep, uh, historically, right? And I want to say what those things are fundamentally different because of why, right? Because if you think about, oh, I'm going to store something for a long, long time, right? And I don't want to go bankrupt, right? By either paying those guys for, you know, gigabytes of, logs or metrics, so trace tofe, right? Or even myself buying terabytes of storage, right? They have to limit that some way, There is also some limitations because even if eBPF, instrument in everything costs a lot of, uh, you know, CPU, At the same time, if I am interactively diagnosing something, I kind of like enable the bug mode for everything, right? Or at least the things they care and really have some very, very specific, data. to analyze that while the problem is ongoing. And this is why I would say, just saying, oh, you know what, I have a an OpenTelemetry and I feed that to the Datadog, so I don't need any other tools. Right. I don't think that is quite correct.
Viktor
00:28:07.854
I got hooked on when you said the store. Everything forever or something in that direction. how, how do you deal with retention? So I assume that when you said at the very beginning, less is more
Viktor
00:28:25.861
that the definition of less differs greatly depending on the period of time. Like less for last week in my head is very different than less for, um, a year ago.
Peter
00:28:40.801
Yes. Yeah, I mean, uh, if you look at, Coroot. At this point, we have ability to disable some data sources, right, which you don't want. We also have ability to use sampling. for data as well as some other kind of specific, data reduction techniques, right? And then you can, uh, limit the period for which, you store, We do not, do at this point, data, shrinkage in a way. Like, oh, we'll go from five, uh, you know, one second interval to five seconds, or, or like two five minutes or something. Uh, something like that, that is kind of, uh, uh, hasn't been really the, the feature of a product we, uh, work with, right? So typically we would limit that to say, uh, you store the data for a month, uh, in Coroot, right? And that is what is, focused on, right? Again, that's more of a. Diagnostic tool, right? We don't really, design and saying, Hey, you know, I want to really graph solution, like, uh, what's going, what's going on for five years? Right? And to really see, how trends have been operating. That's hasn't been, haven't been a problem.
Viktor
00:29:47.093
Because you know, UN until not long ago, I was very much in favor of. I don't understand. Why would you store data longer than a week a month or something like that? Kind of who cares really about your CPU memory performance networking from a year ago kind of come on, but now I can imagine how that would be useful in some different format. With ai, right? Because there is an obvious usage of ai, Hey, help me fix this issue, right? And then it, uh, digest the data comes up with solution. You say it looks okay, do it, and so on and so forth, right? But I feel that we are entering into the age of when AI could be predicting things like scaling could be a good example. Kind of, Hey, for five years. You had outage around Christmas, you know, because of this and that. This is what we expect based on these trends and stuff like this. So I can easily imagine it's serving double purpose on a, on a very different timescales in a way.
Peter
00:30:55.985
oh. Yes. I mean, I am, uh, I am totally agree with you, right? But I think, uh, this is also an important thing, right? I think if you, are and, uh, uh, relatively early stage startup with a small team, uh, then, uh, it's also very important what you focus on. Right. There is unlimited amount of things what you can do, right. But you just have to choose what you're going to focus on. And this kind of a long-term trend in just was not a focus, focus at this point. I'm not saying it won't be it.
Darin
00:31:25.289
So let's talk about that small. You are referencing yourself with that small startup, but let me flip it around. Let's say you've got a small startup that's wanting to use Coroot. They don't mind self-hosting. That's easy for them. What is, uh, the minimum. Stack for them to get a functional Coroot, not just a hello world, but something that I could throw one or two apps at and it not even Blink.
Peter
00:31:52.509
Oh, I, I mean, uh, if, if you, you're talking about like resource usage. Uh, yeah, I mean, uh, I would, uh, say we did not try to have a, a Coroot to be, you know, like a super. small, uh, resource, uh, usage. But I would say if you have a server with, four, uh, gigabytes, of memory, right for that, uh, that server that can, operate, uh, you know, quite fine for maybe, 10, uh, Node.js, if it's, uh, if it a decent amount of pods.
Darin
00:32:24.642
so that seems reasonable. You say four gig and four gig all of a sudden feels really small nowadays, because it is, but it could still work, is what I'm hearing.
Peter
00:32:33.897
Yeah. Yeah. I mean, I may even install it on the smaller, smaller amounts. But I mean, I would, uh, rather sort of, and the promise not deliver. Right. In this
Darin
00:32:44.330
Yeah. I wanna go back to the top of the show where we're talking about the open core model around Coroot. Why did y'all choose to do that? Does it just feel like it was something that you, that needed to be there and
Peter
00:32:56.337
Yes. Well, what I think, uh, uh, in this case look like, uh, let's go back maybe to the Coroot, Genesis. Coroot was released, uh, started first by, uh, Nikolai Cko. before that he built and, uh, sold the SaaS observability product. Uh, right in this case, which did a kind of a lot of, like a similar things to what Grafana is doing, but kind of like a, uh, simple behavior, going to capture a lot of data, provide a lot of visualization, and so on and so forth, right? wanted to, get this another take, right as saying, Hey, we want to really focus on providing answers, simplifying things, and so on and so forth, right? But also you can think about the observability, market was very mature at this point. All right. Like it's, uh, I just, uh, went to, KubeCon and uh, you would see like took a picture. There's probably like a hundred of observability companies, right? And each kind of time I come to the show, right, there's like a 50 more. It's a rather, crowded market. And, uh, in this case from a business standpoint, I think that is where, open source, uh right, has, uh, ability to, attract people to give a second look Ability to, disrupt things, right? Especially I think if you, if you are looking at the open core business model, and if it's not really cripple wear. If your open source solution is, actually quite, powerful. That can be very, helpful. I think it's also changes, uh, in this case how much, capital that requires, uh, to get, in the market. Because if you think about the, open source, I think there's much more, people are much more open, talking to each other about, uh, tools, kind of, and, helping the projects, right? Which, they've been given. So I think that's an answer. And if you're asking about, oh, but why didn't you do it as a peer play, open source. Well, there is also a need to. Uh, build a sustainable business out of it. We did not want to go somewhere, as you know, some other companies doing. Hey, we guys are, open source forever, and then, oops, we can't make any money doing that, so we are going to change a license. And now everything is not open source. And we found what for observability space in particular, that is a way which allows us to build their, sustainable business, uh, and also, give back to open source community. They're both, Me and then Nikolai own a lot to their, open source community.
Darin
00:35:27.069
Well, I think it's interesting that you brought up the bait and switch. most of the time when you see people that bring something in open core and they make a big deal about it being open source, uh, it's always a GPL or BSL or some other license, whereas you all went. With Apache licensing. Apache two. Oh, to be specific,
Darin
00:35:48.070
did you pause at that or were you just like, look, I, in your background, you come from database and I believe you were at Perona, correct.
Peter
00:35:55.600
Yeah. Well, I mean, I am, uh, the founder of Per Corner. Yes. And I'm still involved with Per Corner company. Yeah,
Darin
00:36:02.751
But again, it was based on Perona. Come on Darren. You can think about this. This is my sequel. Right? Was it based? Based around, so again, let's not talk about Oracle and Oracle things, but you feel much more comfortable staying in what I call the true open source licenses. Yes, ag P'S open source, but it's got a lot of, I can't build a business around it, whereas you took the exact opposite. I could build a business around using Coroot Community Edition.
Peter
00:36:32.661
Oh yeah, absolutely. And I'll tell you more, right? There is, quite a few companies out there, which use, uh, Coroot agent. And then, you know, same as I can say, Hey guys, you can, get the data with, uh, Graf and they utilize their own, gui, right? Some are in the cloud, some are also like self-hosted solutions. and that is, uh, uh, fine, uh, with us. We also get some, uh, good testing from them. Good, contributions back, sometimes in, in our agent, right? So, uh, we decided what that is, the balance, which is going to work in, in this regard. Now, I think if you, uh, our philosophy in this case is this, right? We really want to maximize the adoption. So, hey, we, we don't want to have, like, any restrictions about how Coroot open source can be adopted. Hey guys, you want to adopt part of it. You want to build in Coroot in your, let's say, some commercial offering. You ship to the customers. You can all, do that now. We believe, uh, in this case if had a restricted adoption also, we, have, more adoption. And then, you know what, some companies are going to grow up. Uh, in their complexity, there may be, compliance issues, right? So we'll want to get, uh, our, Coroot, enterprise edition, So that is a long term game for, uh, game for us, And I think there is also like a, being a relatively small startup with a small team, right? We don't have a high bond rate as, uh, some other companies are. That gives us luxury of patients. Which I think a lot of this, uh, you know, high octane and venture funded companies, we just, you know, don't have, they kind of have to really, you know, deliver a lot now. Right. And that is where I think a lot of those bait and switch license changes come from because, well, where, kind of still grow in revenue wherever it just, not the exponential growth. Right. Which was, baked in then, they took some venture capital on very high valuation.
Darin
00:38:38.446
I did not do my math as well as I should, or d done my research. has Coroot taken vc,
Darin
00:38:52.429
Okay. That's, that's good. So that explains a lot because you're not feeling the pressures to AI all the things. I mean, you, you feel pressured to get AI into things, but you're not being forced into, okay, why isn't 90% of this AI right now?
Peter
00:39:07.184
Well, that's right. and I think that is a very, uh, very important, I think that is what is also important for us, uh, as founders, is to take the, uh, really good, uh, decisions which have it, you know, technical merits, right? And not just be. carried away with, I would say like a fla of a day, right? Because always we have been always like a hype cycles kind of coming in, right? Uh, I think this is, uh, here's also something which is important, right? I think it's very important as a founder to think, what you're really best. at Right. And there are some people who are, I don't know let's say, born salesmen they're very, very kind of, quick to monetize any hype out there, right? They would be on it the next day. And if you think about, you know, Nikolai and frankly me, we are not that kind of people, Hey, you know what, being a hype expert is not us, But we can build you a generally good. Observability product. And that is what we're going to do.
Darin
00:40:08.851
In building that, again, talking about the AI part of it, I mean, can you, yes. You could have written all the things to do the RCA root cause analysis work. Has AI really helped you in that? Getting to answers better, quicker,
Peter
00:40:23.861
Well, yeah. So, so here is, uh, I think is interesting, right? Because if you, look at. Ai, I think especially over the last few, few years, AI start to mean LLM, Hey, let's us just go and dump everything, kind of, you know, to ChatGPT and friends and we'll just figure things out, you know, just, from that. But, uh, when we think about that, right, there is a lot of, things which actually. For, benefit from like a systematic analysis, If you think about Coroot, the core things of that is what we can actually build their map of your system. We can see how errors propagate, That is not probabilistic. That is not something which, you know, different LLMs may have a different opinions about, right? You can actually have a form that's kind of very systematically, certain kind of, uh, uh, analysis of mathematic precision. And that is what Coroot does, Let's say, Hey, your problem originates here. and I think, I think that is kind of, you can think about like a stage one now. The second thing as well is to say, oh, that's, uh, uh, that's cool, right? Let's say my latency is stored on my database is stored on Amazon EBS of you know, certain type and my latency is high. Now what can I do about that? Okay, cool. That is where we can, uh, uh, really provide all that kind of context and ask Chad, GP, well, ChatGPT or other LLMs for help. That is very, very good having that all the information. Which they have to come up with a good context. Right. But it's not really so much of a diagnostic where a problem is, but really kind of those kind of more specific things like, okay, why exactly the latency could be high right on EBS volume and what you can do about that. Right? If that makes any, makes any sense.
Darin
00:42:25.886
Oh, it makes a lot of sense 'cause EBS volumes are spawned from the pits of hell, um, because they can be really, really bad. So if I was debating on starting with Coroot today, what is my. Steps. What? What do I need to do? What we already talked about. I need a four gig machine. Okay, we'll probably need a little bit more than that, but I don't need to go out and old school days thinking of Grafana. Prometheus. I don't need to go come up with a five node cluster for elastic search. I don't need to come up with X, Y, Z, all these other things. What does it really look like?
Viktor
00:43:02.181
Uh, now you, the moment you mentioned Elasticsearch, you made it easy. Because nothing can be, nothing can be more demanding than that.
Peter
00:43:10.882
Uh, well look, so there, I, I would say it's like a couple of ways. Like you can, uh, uh, do it, which is an easy one, is that if you have a Docker, for example, you can, uh, get the Docker composed, right? It'll, install the car application itself as well as some, uh, the dependencies such as, click house. The easiest way there, I think is to get started is to deploy Coroot on Kubernetes. If you have that, because if you deploy that, we have a helm chart, you get it deployed, it not just, uh, deploys where, Coroot, the application itself, but it's also kind of, gets all your Node.js instrumented, right? So you can say, oh, boom. Right? I have that visibility five. But I would say if you want to explore, you don't even have to start with, deploying code, right? You can, uh, we have a, demo out there, which can give you a very good feel, right? It's, uh, uh, it's, uh, have, you know, deployed. They're sort of like a standard OpenTelemetry demo. They have this kind of application with some failures being periodically injected out there, and you can, see what Coroot, provides if, uh, and to see if that's, uh, that's interesting.
Darin
00:44:21.415
I wanna go back to the helm chart you just mentioned. Uh, we're now in January, which means we're at T minus 60 ish days before ingress. NGINX dies, um, from support at least. Is ingress NGINX in your helm chart at this point?
Peter
00:44:35.994
Oh, uh, you got there as I understand. it's not, but believe me, if it was right, it won't be by that, uh, by that time.
Darin
00:44:45.079
Okay. I'm just, just checking and maybe not just checking, but it's like, oh, I hadn't thought about that. We should check that. how do you feel about that? You said you were at KubeCon this past year. what did you, what was your takeaway from that when you heard that and it's like, is this the first domino to fall?
Peter
00:45:00.717
Well, in my opinion, uh, in this case, that is always this kind of a challenge, right? in the open source and in, and in terms of like an, uh, ecosystem saying, Hey, you know what, okay, what do we have resources or we do want to maintain, right? And on, uh. And at what conditions? I frankly did not dig into that, that much. What I know in general, kind of me participate in sort of often in database and Kubernetes ecosystem. It's kind of interesting, how different philosophies can go. Right? Kubernetes, especially in its early days, was always kind of very, very quick. You know, introduce something, something doesn't work out, you know, depreciate that kind of move on, and keep moving, keep churning. In database we often have very, very conservative. If, you implemented something, you cannot, depreciate that. And even if nobody uses that, then that's probably going to be like a 10 year, depreciation type cycle, Because database is kind of a very slow moving, in general.
Viktor
00:46:01.481
I would agree except that I wouldn't say use the database word over there. I think that what you're describing is can be applied to, maturity or the, the age. Right. Uh. something similar is happening in a way in Linux kernel, right? because it's been around for a while. Everybody's using it. we don't mess up with it. In the same way as we were messing up with Linux Colonels first 10 years, right? Or even with Kubernetes, everybody's much more cautious with Kubernetes. Now I'm talking about Kubernetes project itself, Simply because, hey, it's been 10 years, we are not on in year two anymore, where we can just introduce random ingress and, and see what happens, right?
Peter
00:46:48.405
Yes. Uh, look, I, I totally understand, right? And I think that, uh, happens with, maturity, right? But I think it's also like if you look at every particular point in time, you can see different communities. They do have a different sort of, let's say, attitude to risk, attitude to saying, Hey, we can innovate and then we can, uh, you know, uh, decom things, Like e even if you, the database world, like I say that and yes, like if you look at, you know, my square pause risk, that's kind of very, conservative. But I remember then Mongo came out. Right. It's, it was decades after our database. It was like, who we are kind of moving many very, very quickly, right? It's, we can, break things, remove things, add things. It's gonna change things, right? it's fine. That's how we roll. Right. And that was very, uh, fine for very developer community, right. Maybe, which was like, whoa. Kind of shockingly, strange for MySQL or Postgres guys.
Darin
00:47:50.121
So we talked about how you could get started. You could. Do it with Docker Compose, you could run it inside Kubernetes. If you already have a cluster, you could just try it out on the demo site. But let's say I, you know, go down the path and I go ahead and self-host it, like I'm, I'm liking it. Why would I wanna switch from self-hosting to the hosted version?
Peter
00:48:10.287
Well, uh, first, well, we don't actually have a hosted version, right? So the choice, what you have is, hey, do you want to, run the enterprise version, Or do you want to stay with community? That is probably your, uh, your question.
Darin
00:48:23.871
Okay. That is the question then. So why, why would I need to switch to enterprise?
Peter
00:48:27.274
so if you, uh, look at the enterprise version, there are a couple of different reasons. One is the features differentiating, so Right. some fellow's ai, analysis, They are, built in the enterprise version, we actually have some AI analysis in the community version, uh, where, it's done kind of slightly different. It's, sends the data to, Coroot cloud if you want to enable that with analysis performed there. right drive and on your local host as it exists in enterprise version. Right. But, uh, mostly I would say the heavy duty, root cause analysis, enterprise feature, right? And the second, uh, piece, which comes from. things like, permission, uh, uh, role-based access, right? That is also, in their enterprise version, something which your larger teams needs, right? So if you are the deploy met and scale, you have a different teams which needs to have a different visibility, in Coroot, right? That's what Enterprise version provides. But I think. Another very important part of a open source, right? When you're deploying any software and that, becomes, uh, mission critical. you often want to thinking about, Like, hey, if it's, uh, if it breaks, who's going to be fixing that, right? Or if there's any security issues in that software and so on and so forth. Right? A lot of large enterprises, they, prefer to have a vendor relationship, or often that is even, uh, uh, required in the internal compliance documents.
Darin
00:49:57.213
Compliance documents. Enterprises cannot exist without legal departments, so therefore we have legal and security. So Coroot can be found at coroot.com. That's C-O-R-O-O t.com and uh, all of Peter's contact information will be down in the episode description. Peter, thanks for being with us today.