Peggy Rayzis discusses GraphQL's excellent developer experience.
Peggy Rayzis is the engineering manager at Apollo, where she leads the developer experience team.
Peggy talks about how Apollo touches every layer of development. There are a lot of ways that you can implement GraphQL in your application. It's incredibly flexible. You can even have GraphQL running entirely on the front-end! Peggy recommends that you implement it in your existing application by creating a GraphQL layer that sits between your front-end and underlying services.
Why would it be worth all of the effort to refactor your application from a REST architecture to a GraphQL one? There are performance benefits from switching to GraphQL, but the main draw is the developer experience. GraphQL is much nicer to work with than REST, no more firing up Postman or console logging to get a peek at what's going on.
To get started with GraphQL, Peggy recommends taking a single endpoint in your application and beginning with a schema, then move on to writing your resolvers, get your server running, and then connect your front-end. If you want to learn on something that isn't your product, then Apollo has excellent documentation that is linked bellow.
Kent C. Dodds: Hello friends, this is your friend Kent C. Dodds and I'm joined by my friend, Peggy. Say, "Hi" Peggy.
Peggy Rayzis: Hey everyone. I'm so excited to be on the podcast today.
Kent C. Dodds: Thank you so much for joining us. So Peggy, your last name is pronounced Rayzis?
Peggy Rayzis: Rayzis, yes. I've had people pronounce it many different ways. Racist, razor. Yeah, but it's ... I know, it's terrible. Rayzis is the correct pronunciation.
Kent C. Dodds: Yeah. Well cool. I'm so happy to have you Peggy. You and I have been buddies since, like, I don't know, for the last three years or more. Time flies.
Peggy Rayzis: Yeah, something like that. I know, right? Like, it doesn't feel like it's been that long. I think the first time we met each other was like lunch in New York City with a bunch of New York City Java script people. It was a fun time.
Kent C. Dodds: Yeah. Good times. And so, I guess last we saw each other, was it at React Amsterdam? That was fun.
Peggy Rayzis: Yes, it was.
Kent C. Dodds: Good times. Cool. So, Peggy, you have been, like, as long as I have known you, you've been really interested in GraphQL and now it's kind of your full time thing as it revolves around this GraphQL thing. So I'm excited to talk with you about that, and specifically Apollo and the cool things that you're all doing there. But before we get into that, I think it would be cool for people to get to know you a little bit. So if you could share with us, just whatever you want to share about yourself. Your technical interest as well personal interests or whatever you want to share with us. Yeah. I'd love to get to know you a little bit.
Peggy Rayzis: Yeah, totally. So hey everyone. My name is Peggy and I am an Engineering Manager at Apollo where I lead the Developer Experience team. So my role is kind of a very interesting mix of engineering, marketing and product. I lead a team of Developer Advocates. We're actually hiring right now, so get in touch if you're interested. Also Technical Writers, Docs, Infrastructure Engineers and a Designer. And it's a lot of fun. We ... pretty much our primary goal is to make sure that everyone is successful and happy and productive with GraphQL and Apollo. So, in a given week I could be speaking at a conference and interacting with developers in person. I could be suggesting a way to improve the documentation, perhaps with interactive components and even tutorials. And really just kind of listening to the feedback that our community members give us on Spectrum and Twitter and online, so ... It's a really, really fun job. I get to meet a lot of interesting people, and I get to spread the good word about GraphQL, which I think is something that's pretty much revolutionizing how we're building apps today. So I'm super excited to talk with you about it.
Kent C. Dodds: Yeah, totally. I had this workshop recently, Build React JS Applications where I just take little, like, we have a real world [inaudible 00:03:03] application and we build little parts of it as we go. And preparing that I wanted to go with Graph QL, but I got a lot of feedback from people that, there's just so many who are not in the wonderful world of GraphQL yet that, like, the goal of the workshop was, let's learn React, but considering there a lot of people who aren't familiar with GraphQL yet, if I used Graph QL a lot of people felt like it would actually be a Learn GraphQL workshop, because it's pretty fundamentally different from what a lot of people are experienced with. Is that a challenge that you find? Like, that it's a revolutionary chance and it really is an amazing thing, but has it been a challenge for lots of companies to adopt and if so, what makes it challenging?
Peggy Rayzis: Yeah, so I think there's definitely a learning curve with GraphQL. Like, I'll totally give you that. There was even a learning curve for myself when I first started. I was in major league soccer and we were kind of converting everything over from [Res 00:04:04] to GraphQL, and it took me like a couple weeks to wrap my head around it. And I think it's because it's kind of truly full stack, right? Like you have client portion that you kind of have to understand, which, if you have experience with React you can wrap your head around it, but then, also building the GraphQL layer itself. You have to stand up a GraphQL server, you have to learn, what is a schema? Like, how should I model my data? How do I write resolvers? Then if you're implementing in the production environment you have to consider things like Auth and caching, security, all of that. So when I teach it to people, we used to do the workshop, it was one day. But we've since expanded it to two days because it's just a lot to kind of fit it all and get people really comfortable with running it in production. That's to say, I think the benefits are huge. The developer experience benefits are really ... the ones that won me over from the start and the ones that keep continuing to get better and better, and we haven't really seen too much in terms of push back with adoption. There's lots of companies out there, Airbnb, Expedia, Audi, TicketMaster, GateHub. Like, the list goes on and on, everyone's starting to adopt it, but it definitely takes a lot. And I think you made the right call with your React workshop, not trying to cram in GraphQL, too, because it could definitely get overwhelming for people that are just starting to learn React.
Kent C. Dodds: Yeah, it sounds like GraphQL is kind of ... it's such a revolutionary thing that it touches every side of development. All the way, really far back ... in some cases there are providers who even take care of the database side for you. Like, you get very far in the back, and then all the way in the front, as well, with now you're doing estate management and caching and all of that. And authentication. And so, it touches so many places that it can be a pretty impactful change. It's not something that you can iterate to super easily. And so, I'm finding that quick path to get something implemented so that you can iterate toward making the bigger impact. That can be a little bit of a challenge. One thing I want to ask you about is a couple years ago I was on ... I was running the Java script error podcast, and we had Lee Byron, creator of GraphQL, on the show, and he mentioned something to me that I thought was kind of interesting. I want to get your take on it, was you can actually implement GraphQL on the front end and have the entire thing just running in the front end. And like most of your code, it's treating it like a regular GraphQL server, but all your resolvers and everything are hitting the rest end points that you have today.
Peggy Rayzis: Yeah.
Kent C. Dodds: Have you ever seen anybody do something like that, and would that make sense, as a maybe more iterative, like, less company-wide impactful way to start adopting GraphQL?
Peggy Rayzis: Yeah so there's maybe like two ways you could approach this. I do think, you know, it is an interesting way to kind of get your feet wet, but not necessarily something that we would recommend most people doing in production. So, the first way is, you could literally run a GraphQL server on the client, but the caveat with that is the GraphQL JS modules, in order for validation and all of that, they're extremely large and would inflate your bundle size. So that's kind of why we keep that complexity on the server instead of on the client.
Kent C. Dodds: Right.
Peggy Rayzis: But, I have seen actually teams running GraphQL on low memory devices actually include the server on the client side, as a way to kind of just save bytes over the wire, which I thought was kind of interesting. But the second way that you could do this would actually be with Apollo Local State. Apollo actually gives you the ability to specify client side resolvers. So, in those resolvers you can literally do whatever you want. You could hit a rest end point, you could query a device API, you could augment some server data with Local State, for example. And that could also be a way for you to get the hang of what is a resolver? How do I make queries? Connect them to components? That sort of thing. And people do that extremely successfully. I've seen ... I think I was talking to someone who built the Microsoft Teams app, and they've actually just completely ran GraphQL on the client using Apollo local state. I know people at Sony do that, as well. So yeah, it's definitly possible, but I do think you lose some of the benefits of GraphQL. For example, if you have multiple clients querying the same API it kind of makes sense to shift that to the server so that way you're not duplicating code across your clients. You also lose some of the benefits with smaller payloads, fewer round trips if you're doing it that way as well. But it's still ... I think it just speaks to the flexibility of GraphQL that you can kind of use it in all these very interesting ways. And I think we're kind of only just scratching the surface on what you can do with it, and that's what really make it such an exciting space to work in. Is, it's so flexible you can really mold it and shape into however your organization needs it, which is pretty cool.
Kent C. Dodds: Absolutely. I know when I was at PayPal we wrote ... at PayPal you have the front end, and then there's this like, middle tier sort of thing, that serves the front end. They kind of consider it the front end, but it's actually a node server that kind of hits the end points and stuff. And then we have a technical mid tier, and then we have lower downstream CE and Java services and stuff. And when we were trying to implement GraphQL we had just implemented it in our node layer and the resolvers were all just hitting rest end points. But because it was running on the server, it was not much worse than hitting the regular rest end points and having one off end points. In fact, it was much better. And so I agree with you, I think are lots of different ways to implement this. And I think one of the struggles that I have witnesses in trying to get companies to implement GraphQL myself, is the backend engineers are typically like, "Ah, I'm pretty happy with Res. We have all of these frameworks already that I've been using for years. I know how to do Res-full stuff". And so, it's been difficult to convince them, but if you can stand up like a nodes server that hits their end points ... and then eventually they come around because they're like, "Oh, you mean I don't have to version my end points anymore?" Or, "Oh, I don't have to make a one-off end point for every single little thing that you need? Yeah, that sounds great." They come around eventually.
Peggy Rayzis: Yeah, that's actually our recommended architecture, is really just like, a layer over your existing end points. An Apollo server makes this pretty easy. We have this thing that's called a rest data source, and it very easily Hooks into the rest end points you already have. It even stands up a resource cache for you, so that way your backend engineers aren't worried about increased traffic or possibly taking down one of the end points. So, yeah, I think that's a totally valid path and really kind of how we see most teams implementing GraphQL. It's like, this layer in between their front end and their underlying services.
Kent C. Dodds: Is that mostly because that's the [inaudible 00:11:42] architecture? Or is there something about that ... like, I would ... my assumption would be that the best architecture would be the back end services are all written just with exposing GraphQL end points, and then you have multiple services, whatever you can pose those GraphQL end points in is magic and it's awesome. Or is there something to be said for downstream services, they expose just regular cred end points and then we have this thing in the middle to put it all together. Yeah, I'm mostly interested, why is that the recommended architecture?
Peggy Rayzis: Yeah, totally. So GraphQL native, kind of like GraphQL's native services, I think there's a lot of people doing really interesting stuff in this space right now. A lot of the database providers, like Prisma and [Thesaura 00:12:33]. But, we're kind of not really taking a stance on those. We're kind of more focusing on the GraphQL layer as opposed to GraphQL native services, because, like you said, a lot of back end teams don't want to rewrite all their services to GraphQL and if they do they certainly don't want to use Java script-
Kent C. Dodds: That's fair.
Peggy Rayzis: Yeah, so that's kind of really what we see a lot of times. But I definitly think there's still a place in the ecosystem for those. But ... so as part of that kind of GraphQL layer, you touched upon this a little bit, what we really recommend is not like a monolithic GraphQL layer. We actually recommend that teams stand up a lot of ... kind of take responsibility for their own domain schema, and that way they have multiple sub-schemas that then they've federated into a single layer. So, we're actually launching this ... oh, I don't know if I'm allowed to say exactly when, but within the next couple weeks. It's this essentially new programming model. It's on top of GraphQL. It's completely spec compliant, but it's just a really interesting way to essentially compose multiple sub-schemas into a single graph where the execution is actually distributed. So you know, at PayPal, for example, I'm trying to think of separate teams. Maybe one team has- [crosstalk 00:13:56] yeah what are some, I guess, schemas or teams at PayPal that had GraphQL services?
Kent C. Dodds: Yeah, so the Merchant Team has a lot of GraphQL stuff, so their payment methods, their credit card or bank, or whatever, and then you have your contacts on the consumer side where the people that I send money to regularly ... so is that helpful?
Peggy Rayzis: Yeah, yeah, exactly. So maybe you have this user query, for example, and you have the contacts coming from one service, and you have, you know, the payment methods coming from one. I'm trying to think of a type where both of those would be combined, but for example you want to just be able to seamlessly query the graph, and you don't want to have to think about which services these queries are pulling from. You just want to write a query and have that data come back to you on the client. And that's kind of really what this, the federation solves, is implementing GraphQL, specifically in a micro-services architecture. So that way, it's opaque on the client. You're just requesting the data that you want, and then the distributed execution will kind of take care of sending that parts of the query to the different services that request it.
Kent C. Dodds: Oh that's awesome. Yeah, that's ... so, that was something actually that we were trying to implement before I left. I wasn't super involved in the GraphQL stuff, but I was sort of involved. And that was something that was kind of the goal, was every one of these services ... we not wanted to have some sort of wrapper that exposed a GraphQL end point for that, and then we'd have just one end point that was like Paypal.com/GraphQL and you could just compose them all together in a beautiful way. I'm not sure how far they've gotten on that so far, but I like that idea a lot. That makes a lot of sense. So, kinda cool. So if I wanted to ... I don't know, like ... one thing that might be useful, and I don't want to get too deep into this because there are about a million of these types of ... lots of content on why GraphQL is so great, why we should top using Res, all this. But maybe like a high level, why is it worth all of the trouble to go through this and to make this pretty fundamental shift from our Res stuff to GraphQL?
Peggy Rayzis: Yeah, totally. So this is kind of the major thesis of my React Amsterdam talk, because I think when a lot of people talk, like, why GraphQL? They talk about some of the technical benefits. They say, "Oh, you know, it prevents over fetching and fewer round trips and smaller payloads." And while that's all well and good and I think that's great and it's awesome that GraphQL gives us these benefits, the actual main benefit that you get from switching to GraphQL is really the developer experience. And I think, with that come increased productivity. Teams can ship products faster. They can, you know, hire the best engineers to their teams. They can retain them once they're there because they're happy and they're productive. It's really just, I think, the amazing developer experience that it gives you. So some of the things that I kind of covered in my talk and the demos, because of GraphQL is static typing, it really unlocks the door to a lot of really cool stuff. So, we have a VS code extension, for example, and right inside your editor, as you're typing queries you auto-complete, you can peek inside your schema, you have, if you connect it to our cloud services you have inline performance metrics of how long it takes each field to resolve, right inside your editor. Which is pretty amazing, because if you're working with Res for example, and you want to know, okay what data's available to me? What's the shape of my data? You're either console logging, you're opening up Post Man, or you're tapping the back end engineer on the shoulder and asking them. To just really have all of this at our fingertips within our editor is truly, to me, an amazing thing. So that's one of them. Also, the integration with typescript. Like, I'm super passionate, I guess? I don't know. I think typescript is really the future. I think maybe for smaller things people say it could be overkill, and I think there's definitly a learning curve there, but in a production environment typescript is definitly the way to go. And the Apollo CLI, it will literally read your schema and automatically output type definitions for you that change as you're writing your queries. So now, like, wow! You have fully typed React components automatically, just by running a single command, which is pretty dope.
Kent C. Dodds: Yeah.
Peggy Rayzis: Yeah, and then, like, the local state management experience to me is also something that's super cool, because when you write your React components, they're made up of a lot of different types of data, right? You have some remote data coming from somewhere. You might have some local data. Maybe information about the device. Whether or not it's connected. Information about maybe ... you're building an e-commerce app and you have products, whether or not they're in a cart. And being able to seamlessly mix your remote and your local state in one query is pretty amazing. And even your local state will be automatically typed. You'll have all the same benefits in your editor that you do with your remote date. So yeah. So I think the developer experience to me is the reason why you should do this. It's definitly a shift for a lot of organizations, but it's one that we've seen time and time again, has just been dividends in the companies that have chosen to adopt it.
Kent C. Dodds: Yeah. It sounds so great. And so much of this stuff that I'm doing on a regular basis on my own is ... I'm trying to make a little workshop app, and I don't want to bring in typescript or all of this other stuff, because then no only do they have to learn what we're talking about in the workshop, but also they have to learn typescript. And so I don't get to spend so much time in those technologies, and it makes me super envious to have those sweet editor plugins and things. But it sounds like a really awesome enhancement to your productivity. Eventually we're just not going to code anymore and the computers will just code for us because it's all [inaudible 00:20:26] so perfect. So Peggy, one other question that I wanted to ask about is, most of my audience are probably React developers, but I think a lot of people who follow me online are either React developers and something else, or maybe something else entirely. And I'm kind of curious, I talk a lot about React and it's my love at the moment, but what about people who are not using React? Can they use Apollo or something like it with GraphQL to get some of these same benefits? And if that's the case, how has that changed the story for them, whether they're using View or Angular or something like that?
Peggy Rayzis: Yeah, absolutely. So the really cool thing about Apollo Client is it's totally view layer agnostic. So the core Apollo Client is just Java script. Literally, even if you're not using any view layer whatsoever, you can still make queries and get all the same caching benefits. On top of that are the view layer integration. So React Apollo would be how you use Apollo Client with React. We also have tons of community maintained view layer integrations. So Apollo Angular is awesome. I have not used Angular since Angular One, so can't say too much about the experience there, but the maintainer, Camille, has done a really excellent job kind of integrating Angular into the Apollo stack. And then also, View Apollo, as well is super popular. The maintainer of that, [Geyom 00:21:57] he is on the View Core team, and he's done some just really amazing work with the View CLIUI. How you can kind of just create ... I guess it's like a GUI for a ... a [gewy 00:22:10] for creating View projects, but it has all these really amazing integrations with that as well, and just the developer experience is phenomenal there, so ... I'm not an Angular or View developer, but I've just heard from so many people how it's helped their team. I know they use View Apollo at GitLab, for example. I think one of the engineers there, she just wrote a really great post on how they were able to completely do away with View X after switching to Apollo Client in the local state management feature. So I think it's working just as well for View and Angular and Vanilla Java script developers, just as well as it for React developers. And I think it's really awesome that we've kind of been able to create something that, you know, every developer can take advantage of. Not just the ones who are on the trendy React stack, or whatever.
Kent C. Dodds: Yeah, yeah, right. Yeah, speaking of that trendy React stack, there are some really cool things that have recently happened in React world and also are going to happen. Do you want to talk a little bit about React Hooks with Apollo and then Suspense? My intuition is that, sure we can do some really cool things with React Hooks, but once Suspense lands and is official, blessed for data fetching, then that's when really cool things can start happening. But yeah, do you want to speak to that at all?
Peggy Rayzis: Yeah, totally. So, we actually just released our beta of the new Hooks API for Apollo Client last week. So, I haven't spent too much time with it yet. I literally just converted my demo app to Hooks and it was only two Hooks. But people tell me it's really great, because the API was actually based on a community plugin that we had that became really popular, so we borrowed a lot of the ideas and kind of merged them into Core. And you know, it's just amazing. I think it's really going to solve the problem of multiple mutations. So like, we had a render prop API. Well, we still have a render prop API, it's still there. But in order to do multiple mutations in a single component you would start to have the classic render prop pyramid of doom. So, now with Hooks you won't have that anymore, which I think will be really great. And people can kind of abstract away ... you know, perhaps you have a complicated login flow that has multiple mutations. Well now you can abstract that away into a Hook, which I think is pretty dope. So I'm super excited about that and I can't wait to get it the communities hands and just see what everyone's going to build with it. I think that's going to be pretty transformative. And then Suspense. So we've actually done a couple demos with Suspense and Apollo Client, and it works pretty well, actually. We haven't built anything officially into Core. Actually last week we had a Apollo day at NYC and Jared Palmer gave a really cool talk about Suspense and Apollo Client. He has a demo that he shared. So, for all intents and purposes it works. Although, you know, I don't know what changed the React team is going to have to Suspense like a couple months from now. And that's kind of why we haven't built anything into the Core. It's like, if the React team says it's not ready then, you know, we don't want to build something and have people adopt it and then it all has to change in a couple months. We definitly want to make sure that whatever API's we are releasing are stable. So I'm excited about the possibilities, because now, you know, instead of handling your loading state from every Hook you can just put those Suspense boundaries in. And almost like your boundaries just kind of handle them in a more top level of the React tree. I think it's going to be great. I'm excited for all the new developments with server-side rendering and, you know, being able to stream and not having to do that double pass render like we do now. I think that's going to be really great for companies that use that, use server side rendering. It'll definitly improve performance and user experience there, so ... I'm excited about the future, but you know, obviously Suspense I think is still kind of under wraps and there's a lot of new developments there. So, as soon as it comes out we'll be ready. It just ... we're kind of like eagerly anticipating the release of that just lik everybody else.
Kent C. Dodds: Yeah, it's definitly going to be ... just like Hooks, it'll be another transformative improvement to the framework. And honestly remarkable that it's totally backward compatible and all that, too. Pretty cool stuff. And I think that in particular with something like Apollo, it's very applicable in changing the way that you're going to be using things in a wonderful way. So, as we kind of come to the close of our time here, Peggy, I want to ask, what would ... if somebody is sitting here and they're like, "Yeah, cool, I've heard about GraphQL, I want to give it a try. I love that DX developer experience stuff." What would you give them as some advice of how to get started using GraphQL in their application?
Peggy Rayzis: Yeah, so I think that the first piece of advice that I give everybody is just to start small. Like, literally take one end point in your application, maybe not a mission critical one, maybe something super small, isolated feature, and just start with the schema. So think about the data that you're requesting from the end point on the client. What kind of data do you need to expose? And that's the data that you'll start your schema out with. And don't worry about ... I think a lot of teams start with GraphQL and they want to dump everything into their schema all at once. Don't do that. It should be totally demand driven and only the fields that you're using should be what is represented in your schema. But start there. So you start with your schema. Then next build your resolvers. We have the Apollo Data source package that makes it really easy to kind of hook up a rest end point to your resolvers. You could start with that, or you could even start with plain fetch. Like, that's totally cool, too. And then once you have your GraphQL server up and running, then just connect it to the front end. Hopefully with Apollo. You know, integrate that use query Hook and end query that your new API and ... just starting small with that one isolated feature and building it from the front end to the back end will kind of give you a taste of what it's like to work with GraphQL at all levels of the stack. And it'll allow you to evaluate whether or not it's truly for you. Some people ... you know, I think most people, they kind of see it come full circle and they really ... the light bulb clicks, and they're like, "Okay, now I understand why this is so awesome and we want to roll it out across our entire org." But yeah, just start small. One isolated feature. Build out the backend first, then connect it to the front end and ... yeah. Hopefully from there you'll want to connect a lot more end points, but I think just starting small and keeping it super scoped is how I would recommend most teams to get started.
Kent C. Dodds: Yeah. That makes a lot of sense. Now if there's something that, like, maybe somebody is worried about, like ... often when we're learning new things, I find that it can be valuable to learn those things somewhere else. Not in the product, so that I can make all my mistakes in a place that I can throw away. Are there any resources that you can give our listeners for how they can get started with GraphQL in a place that's not their product, so they can make all their mistakes first and then take all their learnings to their product?
Peggy Rayzis: Yeah, absolutely. So we have a tutorial in our Docs. If you go to Apollographql.com/docs it's like the first one on the left hand side. But, yeah, that really just walks you through building a full stack app with Apollo. I will give a caveat that it's a bit beyond Hello World. So, this is something that kind of is more ... we built it in order to get people up to speed with like, say they want to create a GraphQL POC at their company. This tutorial is going to gave them the skills they need to do that. So, it goes over things like authentication. It goes over pagination. Local state management. Queries, mutations, pretty much every thing you need to know to get that first POC going. So, it'll probably take you about two hours to complete it, but once you do, it kind of sets you up for success and teaches you the best practices and the things that you need to know to get going. Building more tutorials is something that my team is going to be working on in the next couple months. I really want to build a type script tutorial. We've been getting requests for that one a lot. So expect to see kind of more developments in that space and more targeted tutorials as we start to build out the team. But for now that one should kind of teach you everything you need to know to get started.
Kent C. Dodds: Very cool. Awesome. All right. So, with that, I think that's pretty much it. So just to wrap things up, Peggy, what is the best place for people to connect with you and keep up with what you are up to?
Peggy Rayzis: Yeah, definitly Twitter is the best place. I'm pretty bad about answering my email. Yeah. I'm @PeggyRayzis on Twitter. Just feel free. My DM's are open to DM me. Also, posting on Apollo Spectrum is a really great way to get answers because I get a lot of DM's and sometimes I can't get to all of them, so posting your question in the Apollo Spectrum channel is a really great way for community members to help you and hopefully, you know, have a link that can be used by other members in the community who might have the same questions. So yeah, definitly find me either on Twitter or find me there as well.
Kent C. Dodds: Perfect. All right, thank you so much Peggy for all that you do and thank you for giving us a half hour of your time. And with that we will say goodbye. See ya.
Peggy Rayzis: See ya. Thanks for having me.