text
stringlengths
0
2.35k
So the use case you give is one where maybe you have some less-traveled parts of an API, and all of a sudden you think "Okay, can I or can I not change this, or make some modification tweaks, whatever it is, maybe even make some backwards-incompatible changes?" So what you're saying is that you want to have enough info...
**Anthony Alaribe:** I mean, very much. Of course, within the limits of what a machine can learn about your servers; it tries to understand "These are the fields that are usually being sent, this is the frequency at which they're being sent." This particular field, this name is usually - maybe it's a string, it's a tex...
**Johnny Boursiquot:** So is this a development time tool? You want this detection before going live with a change, right?
**Anthony Alaribe:** Yeah. In production, in real-time...
**Johnny Boursiquot:** So before you go production, right? Or when are you detecting -- when are you saving my bacon? \[laughter\]
**Anthony Alaribe:** I mean, Smile, you wanted to add something first?
**Smile Egbai:** Ideally, we would all love for this detector to our code before you go to production, and definitely the tool would surely run both on your staging server and your production server, and whatever servers you would have... But it all depends on the tool itself detecting these anomalies. So if you have t...
**Anthony Alaribe:** In a perfect world, a tool like this probably should not exist, because if people were building perfect systems, will test it, every single endpoint is tested, every single field is tested...
**Natalie Pistunovich:** And everybody is following your updates...
**Anthony Alaribe:** Yeah... \[laughs\] Then you really don't need something like this. But unfortunately, in the real world you always make compromises. People deploy to production on Friday nights... And the issues don't only come from within. So you have a service - the issue might not be that on the server side thi...
\[12:11\] For example, if a consumer used to send a particular format of input, and someone deployed a web app or a mobile app that starts sending something else, then we also want to flag that and notify someone that "Hey, this server is accepting a very different input from its clients." And the clients could be thir...
**Smile Egbai:** Something that came up a lot in our interviews, where -- it's actually based on third-parties; we integrated a third-party provider. And I think it's also a bad practice when you version an API and you end up going back to version one and changing things. Meanwhile we know if you are going to be using ...
**Anthony Alaribe:** I actually just remembered a conversation we had in the interviews with someone, and he said that they have a popular fintech provider that they consume/make use of, and this fintech provider probably deployed a change, and everything was broken for about a couple of hours. So they lost a lot of mo...
If you have something that shows you and says "Hey, this contract broke at this time, and this is what changed", you can show and say "Hey, you definitely broke something. This is what you broke, at this time."
**Break:** \[14:47\]
**Natalie Pistunovich:** The first time that we talked about APIToolkit, I kind of imagined myself that a good scenario for me to use that would be as a consumer of different APIs... But just now y'all mentioned that this can be a great tool also for a provider of APIs, to make sure that all are in sync. So that's a ve...
**Johnny Boursiquot:** So along those lines... So it sounds like - and we're gonna get to where Go fits into all of this in a bit here... I wanna make sure I understand your project before we dive in deeper. So your website APIToolkit.io mentions observability. So observability means something very specific these days....
And then you have tools, whether it be on the provider side or on the client side, which obviously APIToolkit.io could be something that you as a consumer or me as a buyer, the consumer of this API, could use to basically create my own SLOs, around the availability of this API that I'm paying for and I'm consuming.
So what does observability look like for this project? Are you providing tools to track over time what the performance is? What kind of information are you surfacing with each tracking that you're doing?
**Anthony Alaribe:** It's funny that you say observability is very specific these days. I actually think of observability as a word is just -- people are using it to mean so many different things... Probably APIToolkit is joining that list. But when we say observability in the APIToolkit context, we actually mean what ...
**Johnny Boursiquot:** Is your focus on the shape of API requests and responses? We've come back to the field and the endpoints a few times; I wanna make sure that when we talk about observability, we're sort of localizing the word... We're contextualizing it to just basically the shape of the API; less so things like ...
**Anthony Alaribe:** \[20:01\] Yes. I mean, we give you your throughput, response time, and different percentiles. That's kind of like just the cherry on the cake; we have the information, so why not give it? But the exciting part of APIToolkit is that you get these statistics, but down to the level of the fields and t...
**Johnny Boursiquot:** Okay. But you have a few layers on top of that, in addition to just the fields stuff.
**Anthony Alaribe:** Yeah.
**Johnny Boursiquot:** So I do want to sort of dig into the Go side of things... What role is Go playing in this product? I know you have a lot of plugins and integrations sort of a - and go is obviously one of them - is go playing a bigger role in this project for you?
**Anthony Alaribe:** In the development world right now, if you want to do a lot of numbers crunching, and handle/process a lot of things in little time, you have a lot of options. But in terms of the popular options, which are also easy to get into, if you're gonna on-board a lot of people on the project, I think Go b...
**Johnny Boursiquot:** Can you think of ways that Go has helped you solve a problem that perhaps would have been more difficult with a different language? Or did you even consider any other language? Have you used any other language on this project?
**Anthony Alaribe:** Our stack is actually -- we have clients for multiple programming languages, but in terms of the backend side, we mostly have Go and Haskell.
**Johnny Boursiquot:** Wow, Go and Haskell. That's an interesting pairing. Tell me more, tell me more.
**Natalie Pistunovich:** Which one is the frontend? \[laughter\]
**Anthony Alaribe:** A lot of the frontend stuff is actually Haskell. We use Haskell for things where performance doesn't matter so much. But one of the reasons why we use Haskell is that for certain kinds of problems, it can be a little more precise to represent. I mean, you can think of APIToolkit as a kind of parser...
**Johnny Boursiquot:** So I'm curious where your toolkit lives. Are clients routing requests through it for this analytics to be captured, or are you at the edge, at the API gateway level, or where is this deployed?
**Anthony Alaribe:** \[24:07\] So there's the ideal scenario, there's the long-term scenario, and there's the now.
**Johnny Boursiquot:** Okay, let's do it. \[laughs\]
**Anthony Alaribe:** In the now, what we have are language middleware. For example, in Golang you install middleware, depending on -- if you have a web service, you install middleware for whatever Golang router you use. And what this middleware does is it basically takes a copy of your requests and strips out sensitive...
You can sample the request, but we basically process every request that we get on the backend to make sure that that request looks like what the server was expecting. So that that request actually fits the contract that we have. So it's a system that's continuously learning. The more requests from a given service we ge...
**Johnny Boursiquot:** Okay.
**Smile Egbai:** Think of it as a lawyer...
**Anthony Alaribe:** Explain more, please... \[laughs\]
**Johnny Boursiquot:** Explain. \[laughs\]
**Smile Egbai:** You have two clients, and APIToolkit will seem to be the lawyer. One client doesn't have a lawyer, but the other client has a lawyer, so it's just sitting there in the middle, vetting contracts. You could think of it that way.
**Johnny Boursiquot:** Okay, okay. So I can see how Go in this case would be a good choice as middleware, as one of your clients. So if the shuttling of data on request happens every time somebody makes a request, and a goroutine gets launched, and during the processing of the request hopefully anything that you're doi...
**Anthony Alaribe:** This is why I told you there's the now, there's the idea, and there's the future. \[laughs\] So for a lot of projects, for example Golang projects, the middleware app which works very nicely, where the middleware is part of the project; Go's memory representation of information is very compact and ...
So the plan is to have these kinds of sidecars... Actually, we do have the sidecars, but there's just no one using it yet. But the plan is to have this sidecar which you can -- if you have like a Kubernetes cluster or a Docker cluster, instead of your application sending your request to our server directly, it sends it...
**Johnny Boursiquot:** \[27:59\] Okay. This sounds very much like the OpenTelemetry community right now. One of the design approaches is basically to have these collectors that are running, and you can certainly run them as sidecars if you wished... But if you have dedicated collectors where - as the name implies - the...
So what I'm hearing is that you're hoping that even though Go today makes the middleware approach good enough, that's ultimately not the long-term solution, right?
**Anthony Alaribe:** I'm happy you actually mentioned OpenTelemetry... I mean, you can think of the information we're getting as traces; it's basically traces. So the entire system is very much inspired by how OpenTelemetry is designed. And the collector is basically language-specific. So if you have a Golang server, t...
So outside of the collectors and our server side, Go is the first line of contact in terms of processing these traces when they come in. That's where we actually see the real beauty of Go.
**Johnny Boursiquot:** Right. For certain languages in certain tech stacks you're gonna have sort of optimized for those tech stacks sort of collectors and things.
**Anthony Alaribe:** Precisely.