Algora | Algora TV

Open source interviews #27 - Tushar Mathur, founder of Tailcall

Algora

Watching now

00:00 Hey everyone, welcome to the new season of the Open Source Founder Podcast.

00:03 I'm Yanis and joining me today is Tushar, founder and CEO of Tailco.

00:09 Tushar, thank you so much for joining.

00:11 I'm excited to start and would love to hear some information about your personal background

00:15 and how Tailco got started.

00:18 Sure, so thanks for having a little bit about myself.

00:22 I've mostly been on the engineering side,

00:27 you know, hands-on coding, engineering leadership,

00:29 management, those kind of things.

00:33 It's been a year since I started off working as an entrepreneur,

00:35 starting off doing more than just coding.

00:39 And I'll give you some more context

00:41 about what I was doing.

00:42 So I was actually prior to this

00:44 working with a company called Dream11.

00:46 There I was leading engineering.

00:48 I was there for like six years.

00:50 And Dream11 is a very large fantasy gaming platform.

00:54 And when I say large, I mean,

00:55 we used to get a huge number of users.

00:59 And fantasy gaming is basically, you know,

01:00 real sports that is happening on the field.

01:02 You end up creating teams

01:03 and you virtually sort of play with every other folks.

01:06 You create a virtual team and you kind of compete with other people.

01:09 So cricket is very popular in India and people used to just flock onto our platforms, huge, huge amount of scale.

01:15 And that's where we realized, you know, something that we need to invest in our edge there.

01:22 So the edge is something that sits between your mobile app and your backend.

01:27 So I spent a lot of time learning about it,

01:30 going very deep with what kind of optimizations we can do on the edge,

01:34 what kind of workload can we shift on the edge.

01:37 And something that I realized was a lot of companies haven't truly,

01:41 you know, leveraged the power of what edge can do today.

01:47 And it's not something that for a large organization,

01:50 it's a burning issue, but for a small organization,

01:52 it can be quite valuable if you just start early.

01:57 So we started building off Tailcall from all those learnings.

02:00 So Tailcall essentially is an API platform that's edge compatible,

02:04 and you can actually

02:06 use this to power

02:07 all your APIs.

02:08 It becomes like

02:09 a unified entry point

02:10 to access any kind

02:11 of APIs

02:12 and it gives you

02:14 very high performance.

02:15 It does tons

02:16 and tons of optimizations

02:17 that are very difficult

02:18 to do if you

02:18 are handwriting

02:20 all that code

02:21 and,

02:22 yeah,

02:22 it's designed for,

02:23 you know,

02:23 very early stage startups

02:25 to whatever scale you want to use it at.

02:30 Perfect. Thank you.

02:31 Thank you for the introduction.

02:32 And since starting, what has happened

02:36 in terms of the community, in terms of the traction,

02:39 first people using Tailcall?

02:42 So very interesting because the timing of this

02:45 is we

02:47 actually started writing a

02:49 blog post about our

02:51 whole journey in the last one year.

02:53 We open sourced our project with Scala

02:55 and Scala was the stack that was chosen.

02:57 We knew Scala, we had a lot of

02:59 love for the language,

03:01 functional programming and all of that.

03:03 And then we kind of migrated to Rust.

03:05 And, you know, an open source product migrating from one stack to another

03:09 basically generally leaves a bad taste in everybody,

03:13 every contributor's, you know, mouth.

03:14 So what we wanted was kind of timed it properly

03:18 and we talked about the journey in detail,

03:20 but the basic idea was do it open source, leverage the community

03:24 to kind of build

03:25 most of the work get feedback from developers and that's one thing that i've really uh i mean

03:33 your platform has really helped us build the software that it is today because we've been

03:38 able to access a lot of talent a lot of folks who are just passionate working you know day in day

03:43 out weekends nights they don, it doesn't matter.

03:45 They'll just keep creating pull requests and we'll just review them.

03:48 So it's a lot of fun and got a lot of work done for us, right?

03:51 I think this is one of the true ways open source can become sustainable.

03:56 Otherwise, it's very difficult.

03:58 You can open source a repository and for most companies,

04:00 it becomes like a marketing GDM strategy, right?

04:03 Like, okay, we'll open source it.

04:05 Developers will use it, give us feedback, that kind of stuff.

04:08 We are actually getting work done.

04:10 And that's been, you know, one of, I mean,

04:13 a very successful effect of us being open source.

04:16 I would open source everything because it's so easy to get work done.

04:22 One thing that we realized was, you know, we don't have to

04:25 prioritize it. And for a startup, that's insane, right? Because we get tons of ideas,

04:32 we create an issue, and somebody is there to kind of work on it. So a lot of times those ideas are

04:37 half baked, but people kind of build on it and ask the right questions and it just gets done. So that's been, I mean, I think open source is not leveraged

04:47 by most companies today and should be.

04:51 There are a lot of ways you can.

04:53 Absolutely. Fantastic. It's a superpower. Absolutely.

04:57 And, you know, thank you.

04:59 Thank you also for mentioning Algor and I'll try to circle back on that

05:02 because Telco has been the most active project on the platform and certainly want to hear more about your experience.

05:09 In terms of the sustainability of the project and starting with some funding to get the ball

05:14 rolling, would you like to briefly discuss about this? Sure, so we raised some capital from investors and that sort of helped us hire

05:25 a core team.

05:28 But the core team

05:29 actually right now just works on architecture,

05:32 build the framework right, set the

05:34 right tracks

05:36 so that features can be built on it.

05:38 But even right now,

05:39 I think majority of the work is done by

05:41 open source content. So that's

05:43 how it's been.

05:48 And that sort of helps us control our burn rate also, right?

05:53 Because the way this bounty system sort of works is you get paid if you get a pull request merged.

05:57 So you're not paying for, you know, downtimes.

06:00 It turns out to be a lot more efficient from a startup's runway point of view, because you're not spending so much and hiring a lot of engineers and spending time prioritizing, giving them tasks that they can do.

06:13 And, you know, it's much simpler.

06:17 Totally, totally. And through that, you obviously find people to extend the team with.

06:23 And everyone is remote right now right

06:25 or did you ever get together so far? No we've met a couple of them, a couple of

06:32 folks who work with the team but yes everybody is remote and that's the model that sort of

06:39 works for us. Totally and in terms of monetizing tail call and if you could uh describe what that process

06:47 looks like today how do you envision it in the future all right so at this moment we are trying

06:55 to get uh you know distribution right so we want the software to be open source it's free so every

07:02 feature that you we build is available for everybody to use.

07:05 So if you're a large organization or small organization, you can just deploy it on your infrastructure and start using it.

07:10 The way we are thinking about monetization is if you want somebody to manage infrastructure,

07:17 because especially at scale, this can get fairly complex.

07:21 Infrastructure is not something that developers like to deal with. So we

07:26 kind of can help an organization

07:28 manage it for them.

07:29 So that's how we monetize.

07:33 Absolutely.

07:34 Absolutely. And

07:35 based on that too, it will affect also

07:38 how the team shapes up and whether new funding

07:40 is going to be needed. These are all related.

07:42 And in terms of distribution,

07:44 is there any, now that the terms of distribution, is there any,

07:45 now that the topic came up,

07:47 is there any piece of advice

07:48 or maybe any mistake,

07:51 trial and error that you experienced

07:52 you might like to share?

07:53 Because this is a challenge

07:55 for a lot of people.

07:56 And of course,

07:57 it's oftentimes what makes

07:59 or breaks a project.

08:01 So how have you handled that?

08:03 So I think

08:04 the stack really matters when it comes to a product like ours. We started off

08:11 with Scala, which was probably not the best stack for a product like this. One is performance,

08:17 but it's also the way a jar is sort of distributed. It doesn't make a really slick, you can never

08:23 make a slick CLI application using Java.

08:25 It's very hard.

08:26 You can,

08:26 but it's arguably harder than doing it in say Golang or Rust or something

08:31 that combines to a native binary,

08:34 right?

08:35 So getting that piece is very important.

08:37 How,

08:37 how easy or difficult is it to install a product like us?

08:41 And for us,

08:42 what we did was we wrote everything in Rust,

08:45 but you can install it by NPM. And NPM is something that's not designed for us, what we did was we wrote everything in Rust, but you can install

08:46 it by NPM.

08:47 And NPM is something

08:48 that's not designed

08:49 for us,

08:49 but like,

08:50 everybody's got it.

08:51 So you just,

08:52 you know,

08:52 it's easier to kind

08:54 of install.

08:54 Distribution has

08:55 just the idea of

08:57 getting people

08:57 to install it

08:58 is an important

09:00 aspect of distribution.

09:01 And once you install it,

09:03 how do you use it?

09:04 Do you understand the problem well?

09:05 All those kind of things. So we spend a lot of time

09:07 talking about it. We were writing

09:09 a lot of content and

09:11 started to write on Reddit

09:13 and all these other

09:15 channels, writing blogs about it.

09:17 That's how people end up getting

09:19 to know about the problem statement and the solution

09:21 we're getting. So if you're

09:23 an open source developer,

09:26 distribution basically comes through, you know,

09:28 this is essentially marketing,

09:30 right? I know devs don't like marketing

09:32 but this, I mean,

09:33 they'll run away from the term marketing

09:36 but every blog

09:37 is content marketing, right?

09:39 Every time you go to a conference and

09:41 speak about your product, we're essentially marketing

09:43 the product, right? So you've got to do this. If you don't do it your product, we're essentially marketing the product trade.

09:47 So you've got to do this.

09:49 If you don't do it, you never get to know.

09:52 Absolutely, absolutely.

09:55 And I'm taking that piece of advice too.

09:56 We all should do it.

09:58 It might not feel like the most glamorous work,

10:02 although a lot of people do enjoy writing, but yes, 100%. And in terms of launching online and again something that should happen

10:07 recurringly how have you approached this so far and you know anything anything to share

10:14 with other founders so one thing that we stayed away from was you know building a hype so a lot

10:20 of folks okay you know don't talk about the product, keep building a hype.

10:26 Then one day just set up everything, launch it on Product Hunt and launch it there, create a launch blog and stuff like this.

10:35 We never did it this way.

10:37 We were sort of very organic and organically we saw, you know, we just started writing blogs, some bits of it, talking about it in certain places, maybe a couple of posts.

10:46 It was very unorganized, not very structured.

10:49 And I would recommend putting some structure to it, to a process, to GTM essentially.

10:55 But preparing for that big launch is just, you know, futile.

11:01 Like for an open source product,

11:06 you'll get that spike on day one and

11:08 it just drops to zero.

11:10 It's more important to keep improving

11:12 feature by feature, talk about the product,

11:15 build content,

11:16 and slowly keep adding

11:18 users one by one.

11:20 Initially,

11:21 if you're a B2C company,

11:25 you can do an advertisement and you can get a lot of users on your platform.

11:28 You can sell stuff for free

11:29 and get a lot of users. With open source,

11:32 you can't do that.

11:34 Especially if it's

11:36 a new open source product, you can't

11:37 pay a developer to use your product.

11:40 It's very difficult to

11:41 pull that off, right? So

11:43 literally, for us, we had to talk to individual developers.

11:48 So get one developer to kind of use the product and give us feedback, try it out.

11:53 So that's how we started building.

11:56 And based on that feedback, we optimized, you know, how do we onboard people and stuff like that.

12:01 So if you're an open source company, you'll have to kind of work

12:05 much harder

12:06 because,

12:08 especially if

12:09 you're selling

12:09 dev infrastructure

12:10 kind of products,

12:11 the bar

12:12 for dev

12:12 infrastructure

12:13 is very,

12:13 very high

12:14 because

12:15 you've got

12:16 the zero

12:17 chance of

12:17 failure if

12:17 it's a

12:18 dev

12:18 infrastructure

12:18 product.

12:19 You can't

12:20 be down

12:21 for even

12:21 a second.

12:24 Absolutely.

12:25 And so certainly more challenging

12:27 for infrastructure and developer tools.

12:30 But of course, again, the success scale of it

12:32 mirrors the challenges involved.

12:37 How have you been enjoying working with Rust

12:40 and coming from the Scala ecosystem?

12:45 It's been a fantastic experience, honestly.

12:48 We've been doing performance engineering for a very long time,

12:53 built a lot of products at scale,

12:55 done a lot of performance optimization on various stacks.

12:59 With Rust, it's been so much easier.

13:04 It's not like in most other stacks

13:06 that we used to work with

13:07 typically JVN,

13:09 Node.js, you know all the other

13:12 garbage collected

13:13 programming languages.

13:15 We were mostly kind of

13:17 tricking the JIT compiler

13:19 figuring out essentially

13:22 there's witchcraft around it and

13:24 sometimes the JIT will optimize, sometimes it won't.

13:27 So if you know exactly when it's going to,

13:29 so you write code in that particular fashion, right?

13:32 And what we used to realize is,

13:35 yes, you can write JavaScript as fast as C

13:38 or Java or Scala as fast as C probably,

13:42 but it's unmaintained.

13:43 It's very difficult to keep adding features

13:46 and maintain performance because you'll make some refactor and suddenly your performance

13:53 has become half. And you don't know why because it just looks logically the same. I just moved

14:00 this to a different class. Why should it be slower? But yeah, now the JIT is not inlining that piece

14:06 of code, which it was.

14:07 These kind of things end up making a

14:10 big impact in terms

14:12 of performance.

14:15 It's harder

14:16 to write high

14:18 performance application because it

14:19 doesn't scale. It keeps breaking

14:22 features. Every feature you just end up

14:24 losing performance. 10%, 20%. End of year you're just you with rust it was amazing

14:30 because it was very clear that this means of course we optimized if you

14:34 optimize what kind of what are we saving on and if you add if you make some

14:39 refactors what is it exactly that is going to happen behind the scenes we are

14:43 locating more memory if you're locating more maybe it is going to happen behind the scenes? Are we allocating more memory? If you're allocating more memory,

14:45 you can look at the board and you can say,

14:47 okay, this is going to slow it down.

14:48 It's not witchcraft.

14:49 It's very crystal clear by just looking at the book

14:53 that what kind of refactors will have

14:56 what kind of impact on the performance.

14:58 So that makes it much easier to write hypercon systems.

15:03 And we have actually been able to write

15:06 our application.

15:08 So we use Async GraphQL

15:09 as our background right now, which is

15:12 a Rust library for running GraphQL

15:14 queries. And we've been

15:16 able to run it faster than

15:18 Async GraphQL in its own benchmarks.

15:20 So we pulled out

15:22 certain pieces from the library

15:23 and it's not just this library there are a lot of other libraries

15:26 that we use and we kind of patch

15:28 certain things and we're able to push

15:30 the limits of

15:32 what the machine can do

15:35 so with Rust it's been

15:36 a fantastic experience from a performance

15:38 standpoint and

15:40 yes I think the syntax is a little

15:42 too ceremonious

15:43 I don't think there's a letter on the keyboard that you don't type when you're defining a function.

15:51 Like literally every key is being pressed.

15:54 All the angle brackets, your colons, semicolons, everything is just utilized.

15:59 All your keys are utilized when you are writing in Rastor.

16:03 At the end of the day it's the focus is

16:05 not you know the ergonomics of the language especially coming from scala which is a very

16:11 beautiful language very elegant very stylish uh rust is a little it's not a good looking you know

16:20 i used to play around with just the fonts uh when was coding in Scala because the code looked so good.

16:26 You know, I want to play around with different themes and fonts

16:28 and just look at it for a while.

16:30 With Rust, I never felt that way.

16:31 I'm like, okay, let's just run this code, see if it's fast.

16:35 That's what it is.

16:38 Thanks for sharing all this.

16:40 Anything that gave you a hard time?

16:42 And then in terms of people contributing to the open source project,

16:45 what's there to be said about using Rust and people's appetite to work with it?

16:53 So Rust doesn't have quite a lot of features that Scala does.

16:58 It's a form of advanced language, a lot of features.

17:03 You can do very powerful function with Scala.

17:07 Lot of abstractions that exist in Scala.

17:11 I mean, you are able to define abstractions in Scala that are horribly difficult to do

17:16 in Rastor, even if it's macros and, you know, nasty things to do it.

17:20 So that's one piece that I'd miss from XCAD. Apart from that, I think once you start getting

17:30 used to it, there are different ways, like Rust doesn't want you to do a lot of function.

17:35 In fact, everything is mutable, but it's compile-time checked. So that adds a little

17:42 bit of confidence that it's mutable but it's compile time check so

17:46 I don't need to worry about it.

17:48 But the idea that it's mutable

17:50 initially is, you know, all

17:51 collections are mutable by design

17:54 and you kind of have to get used to it.

17:57 The contribution

17:58 contributors

17:59 has been actually truly

18:02 amazing. One thing that we realized was

18:03 there were a bunch of people

18:05 who were just passionate about writing Rust.

18:09 They just want to code Rust.

18:10 It doesn't matter what they're coding for, what we're building.

18:14 If it's Rust, I want to be part of it.

18:17 That's the mindset people had.

18:20 And people are very, very excited about open source and Rust.

18:26 had and people are very very excited about open source and REST these two things so for us

18:32 getting contributors in fact hiring has not been a challenge we need to attract very smart people which has been part of our core team now and it's been a challenge forever for me always

18:41 right like every time when I want good talent I I've never seen an engineer saying, okay,

18:46 I want to write the most, I just want to write

18:47 JavaScript. Nobody says that.

18:50 So, or I

18:52 just want to, I just love JavaScript. If I

18:53 could just code more in Java. I've never

18:56 heard anybody say that.

18:58 So, but with Rust, that's

19:00 not the case. We've been able to attract good

19:02 talent as a part of our core team

19:03 and contributors,

19:05 a ton of contributors. It's barely been like two months since we open sourced the Rust

19:10 solution. We've got 45 contributions already and a lot of them are just learning Rust.

19:16 They're just newcomers, they're learning, they're passionate about the language and they end up

19:20 treating it well. That's phenomenal.

19:25 I want to quickly circle back on something you mentioned earlier,

19:29 basically about Algorand, about bounties.

19:32 I'm curious what made you decide to start using it

19:36 and then if there's any piece of advice for other people

19:38 because the scale at which you're using them is huge

19:42 and it's working super well.

19:45 You even had Kirill join the team

19:46 who started as a contributor.

19:49 And so what was originally your impetus to use it?

19:53 And then has the experience been any advice?

19:57 Yeah.

19:58 I think if you can afford to pay contributors,

20:03 you should do that.

20:07 There's a lot more motivation, there are a lot of folks who are interested in doing open source

20:13 and they don't know which product or project they should contribute to.

20:17 So as soon as you add an incentive you will get a lot of interest. Now the quantum of the

20:26 Now the quantum of the amount doesn't really matter.

20:29 So you could just put $10.

20:38 It still automatically becomes tier one issue because most open source libraries have issues which have no bottom.

20:42 If you put a little bit of money, there are a lot of people who are interested.

20:46 And that $10 is not going to make a big difference to anybody,

20:49 but it just helps them prioritize.

20:53 If I can do this, I might as well just contribute and make some money.

20:55 That makes a big difference.

20:59 So we saw, of course, your project needs to be interesting,

20:59 all those kinds of things. Of course.

21:01 But if you can incentivize people,

21:07 the motivation level is phenomenally high

21:09 it's not the same as

21:11 a typical open source project

21:13 which doesn't have any

21:14 for us every issue is

21:19 has a bounty

21:21 and we encourage not just

21:23 the contributors but even internally our own team to pick up any issue which has a bounty. And we encourage not just the contributors, but even internally

21:25 our own team

21:26 to pick up any issue

21:27 which has a bounty on them

21:28 and just claim it.

21:29 It's open for all.

21:30 It's not just for

21:31 contributors outside the company.

21:36 On our side,

21:38 it's been a privilege

21:39 serving you.

21:41 And thank you so much

21:43 for the continuous partnership.

21:44 And I hope we continue delivering this value and learning from you. And, you know, thank you so much for the continuous partnership and I hope we continue

21:45 delivering this value

21:47 and learning from you,

21:48 as I said,

21:49 you're sort of like

21:50 a shining example

21:51 on how to use

21:52 paid open source contributions

21:54 and reward the community.

21:56 Before we proceed

21:57 to the last section,

21:58 which is the demo,

22:00 is there any final

22:01 piece of advice

22:02 for other founders

22:03 that you might like to share?

22:08 I think there's a huge ecosystem of students who are just getting started with programming.

22:19 I think they have no idea of what Alcoda is.

22:24 And I try to talk about it as much as I can.

22:29 I am speaking about your product more than I'm speaking about my product when I'm meeting, you know, folks a lot of times.

22:35 So I think if you can get into that market, it will be a fantastic thing.

22:41 If you go to colleges, especially I know about India.

22:41 fantastic thing. If you go to colleges, especially, I know about India,

22:44 I talked about

22:45 your product in my

22:47 university, and

22:50 there are a lot of other universities just based

22:51 out of India. There's a lot of talent,

22:54 there's a lot of passion around open source,

22:56 but they don't know how to get started.

22:58 And this is a good platform to

22:59 get involved.

23:02 If you can, if you get

23:03 time, travel to these universities,

23:06 talk about the product that you're building,

23:08 I think it can make a big difference.

23:11 Both sides.

23:13 That's excellent.

23:13 I second that, actually.

23:16 That's beautiful. Thank you so much

23:18 for this, and thank you for the

23:20 interview so far.

23:22 If you would like, we can

23:24 jump into a five or ten minute demo of Tailcall.

23:28 Okay, you know show you how quickly Tailcall can bootstrap a GraphQL endpoint on top of rest

23:36 all existing rest endpoints. We can do a lot more than that but just to be within the time frame

23:42 this is going to be a quick demo. imagine you have a set of apis let's

23:45 say these are your apis json base holders you've got closed api let's see your user api so you hit

23:52 post the rest api you get a list of posts each post has a title body you know the user id that

24:00 posted created that post and an id of that post. And similarly, you also have users API, lists, set of users,

24:08 IDs, name, you know, stuff like this.

24:11 There's a relationship between these two APIs and these are the APIs.

24:14 Say user one, just get data about that user one.

24:19 Similarly, you can say posts one and you'll get just the post,

24:24 the post with the ID one you can also do

24:26 something like user id one and you'll get just the posts with the user id one so that's imagine these

24:33 are the two apis that we are sort of playing what i'll do is say what you can just say npm install

24:40 hyphen g and then call the hq table call this will install the call on your system.

24:47 I already have installed.

24:49 This will tell you what is the latest version.

24:53 We are on 0.18.

24:55 Let's do a tail call init.

24:58 Yes.

25:00 Let's say app.graphq app.

25:03 Yes.

25:05 Okay.

25:06 So now you've got these three files.

25:08 This is a YAML file that's useful for the IDE to work.

25:13 This is an auto-generated file

25:15 which has a lot of information

25:16 to make the IntelliSense work really well.

25:19 And this is the main file where we'll start writing our logic.

25:22 So you don't have to write code for this,

25:28 but to expose these APIs as a GraphQL API,

25:30 we'll just start with the GraphQL file.

25:35 So let's say schema, schema has query, these are standard things.

25:37 And let's say type query.

25:42 And type query, let's say, has posts, which is of type post.

25:48 And the way you can query it is, let's say path is posts.

25:50 What I can do is I can set up a base URL over here.

25:53 Base URL.

25:55 And this is actually, you know, IntelliSense is working on this system.

26:02 So you get all the options that are

26:05 available for which are valid options

26:11 just which makes sure that you don't mess around with the DSLs it's there's

26:17 autocomplete it's add post post looks like this oh yeah perfect let me not add user just yet so those also has user id which is

26:33 okay um once you have this let me so this this is the file so what i'm doing over here is

26:40 i'm just saying okay set up a post it uh Use the post API, which is basically on slash post.

26:47 The base URL is specified as JSON base holder.

26:50 And the type post, the structure of the post is sort of shared.

26:56 It's also specified in this file.

26:57 Now what I'll do is I'll just say start the server.

27:00 And then let the posts, posts give you title and query there you go so you've got all your

27:06 posts on the right okay and it made one e-mail call over here which you can see it's a slash

27:11 post and query for post typing body get that id and this is good but this is not true this is not

27:23 the true power of doc hill so what we'll do is we'll add, you know, a field called user or type user and just say

27:31 value dot userid.

27:33 This is just a slight mustache template that we've written.

27:38 We said, okay, the value, this value, the value that is available over here and pick

27:42 the user ID field and make a call here on user

27:47 slash users to get user right when you do that essentially you are you are integrating the user

27:52 information right into the post let's try to start the server okay so the user type was not

28:00 fine so we have to find user okay Okay, cool.

28:06 Let's just drop address for now.

28:10 There you go.

28:11 So, I started

28:12 posts.

28:14 Let's see if we have user, user has name.

28:17 I'll just

28:17 get rid of this.

28:20 So, there you go.

28:22 So, it made

28:22 quite a few requests on the backend,

28:25 you know, figured out all the ways

28:27 because there are hundreds of posts.

28:29 So for each post, it's making a call to get the user

28:31 and ultimately it's able to orchestrate everything for you.

28:34 This is something that's highly unoptimized.

28:38 And that's what you see over here.

28:39 It says there was one N plus one that was found.

28:41 So what you can also do is you can check this file. It'll say check and it'll tell you that there's one N plus one that was found. What you can also do is check this file.

28:46 It will say check and

28:47 tell you that there's one and plus one.

28:49 Get

28:50 the actual information.

28:55 So here we make

28:57 a separate API request for each post

28:59 separately.

29:01 Yes, you're making hundreds of

29:04 API calls. So if there's a

29:05 we're getting a lot

29:06 of requests on

29:07 your GraphQL system,

29:08 you'll end up

29:09 triggering a lot

29:10 of upstream calls,

29:11 which is not

29:11 something,

29:12 which is not

29:13 ideal.

29:14 So you can get

29:15 this information

29:15 in before you

29:16 start the server.

29:17 You don't have

29:18 to actually deploy

29:19 it in production

29:19 and then identify

29:20 that you're not

29:21 getting too many

29:21 requests.

29:22 Before the server

29:23 even starts,

29:24 you can say

29:24 it'll call check and n plus one queries and it'll tell you that there's one n plus one if you go from

29:32 posts to user you know that could trigger an n plus one issue now how do we solve this one option

29:38 is maybe we can add some sort of matching so let's put a

29:42 Maybe we can add some sort of matching. So let's put a match.

29:48 Delay.

29:51 Maybe 100ms delay.

29:54 So what I'm saying is for each, just batch APA calls for 100ms.

30:00 So that way you'll not end up making too many requests.

30:04 Earlier I was making 100 requests

30:06 and even though there were users that were shared across the posts i was still making that request

30:12 again and again for each post with this if i you know let's start the server again let's make a

30:20 request okay so it's a little more optimized right there's just 10 calls over here and we

30:26 didn't have this start the server there were hundreds of calls actually a lot of calls that

30:34 were happening before if you've been able to optimize this way now but if you try to do a check

30:43 the n plus one still remains on this file.

30:46 Even with the N plus one problem is not solved.

30:49 I've been able to sort of batch the API calls upstream, but the N plus one problem still exists.

30:54 To solve that problem, what I can do is I can say, okay, instead of hitting this URL,

30:59 if you remember, I showed you this API.

31:02 If you remember I showed you

31:03 What if we hit this URL?

31:08 this API, what if we hit this URL and what if we say said that you know user ID and

31:11 user ID

31:14 What will happen in that case

31:17 It gives you post for those users

31:23 Similarly, you can use users. You can say ID equals on ID equals to these APIsEST APIs. So you can actually query for data by APIs, by ID.

31:28 Here I'm just getting two users.

31:30 If I could leverage this API, then I could get all the users in one API column.

31:38 And that basically means I need to make this particular URL.

31:43 So I can do that I can just say when he

31:50 He is going to be ID value is going to be okay Matthew is going to be value dot ID

31:59 Why you don't

32:02 So

32:03 Basically the value the user ID is being passed as a query parameter to the user.

32:11 And when you do this, let's see if what check says.

32:19 Check doesn't know if there's anything that has changed. It says that this is still an n plus one. So it needs another

32:25 parameter which is grouped by

32:27 here we just

32:30 say, we say, okay,

32:32 get the response. So this is the last operation

32:33 that we need to perform. So you get, you make the API call,

32:36 get the response and just group

32:38 it by the user ID.

32:40 And when you do that and

32:41 you check, now there's, it says

32:44 at compile time we know.

32:45 Even before we start the server, it says there's no N plus one.

32:50 Awesome.

32:52 Right?

32:53 Let's try to start the server.

32:55 Okay.

32:56 And let's hit this.

32:58 Now we expect to see only two calls, right?

33:01 Exactly.

33:02 So there's a post call and there's a user call with all the user IDs.

33:06 So I just made one call and I was able to get that response.

33:09 And I can batch it over 100 milliseconds.

33:11 So you could be getting thousands of requests on your system.

33:16 It's going to batch all of them together, make one call upstream,

33:19 get the data and send it back.

33:22 So this is just a basic example of what

33:27 a data call can do for you and can go beyond that. A lot of powerful primitives are embedded,

33:34 mostly opt-in because sometimes batching can have, you know, batching is basically making

33:40 you delay for a period of time. So it's not valid. Sometimes you just want to hit those calls and get the data

33:46 and maybe the list is not that long

33:48 and you're fine with, you know, making an N plus one call.

33:52 So N plus one optimizations are often,

33:54 but the good thing is you know upfront

33:56 how many N plus one issues you have in the system.

34:00 So you don't have to actually,

34:01 it can introspect your orchestration spec and identify what kind of investment products can happen.

34:08 So yeah, this, we have quite a lot of features.

34:11 This works on gRPC and you can change the schema.

34:16 We are also working on a way where right now, if you see this, this is all GraphQL.

34:23 The experience of creating data is all GraphQL. We are working on a way where you don't need is all GraphQL, the experience of querying data is all GraphQL.

34:25 We are working in a way where you don't need to use GraphQL.

34:28 What if we expose it as a REST endpoint?

34:31 How would that look?

34:32 So what if you could get that GraphQL experience only within the DSL

34:36 where you specify a type, specify how you want to orchestrate things,

34:40 but execution could happen over REST or gRPC.

34:43 You would get a lot more better performance

34:45 and flexibility.

34:47 So you don't have to migrate everything

34:49 roughly to the sub-product.

34:51 Yeah, that also makes sense.

34:53 I really like how simple everything was