Awesome
<img alt="webrpc" src="https://github.com/webrpc/webrpc/raw/master/webrpc.png" width="1024" />webrpc is a schema-driven approach to writing backend servers for the Web. Write your server's
API interface in a schema format of RIDL or JSON,
and then run webrpc-gen
to generate the networking source code for your server and client apps. From the schema,
webrpc-gen
will generate application base class types/interfaces, JSON encoders, and networking code. In doing
so, it's able to generate fully functioning and typed client libraries to communicate with your server. Enjoy
strongly-typed Web services and never having to write an API client library again.
Under the hood, webrpc is a Web service meta-protocol, schema and code-generator tool for simplifying the development of backend services for modern Web applications.
- Getting started
- Code generators
- Quick example
- Why
- Design / architecture
- Schema
- Development
- Authors
- Credits
- License
Getting started
- Install webrpc-gen
- Write+design a webrpc schema file for your Web service
- Run the code-generator to create your server interface and client, ie.
webrpc-gen -schema=example.ridl -target=golang -pkg=service -server -client -out=./service/proto.gen.go
webrpc-gen -schema=example.ridl -target=typescript -client -out=./web/client.ts
- Implement the handlers for your server -- of course, it can't guess the server logic :)
another option is to copy the hello-webrpc example, and adapt for your own webapp and server.
Btw, check out https://marketplace.visualstudio.com/items?itemName=XanderAppWorks.vscode-webrpc-ridl-syntax for VSCode plugin for RIDL synx highlighting.
Code generators
Generator | Description | Schema | Client | Server |
---|---|---|---|---|
golang | Go 1.16+ | v1 | ✅ | ✅ |
typescript | TypeScript | v1 | ✅ | ✅ |
javascript | JavaScript (ES6) | v1 | ✅ | ✅ |
kotlin | Kotlin (coroutines, moshi, ktor) | v1 | ✅ | |
dart | Dart 3.1+ | v1 | ✅ | |
openapi | OpenAPI 3.x (Swagger) | v1 | ✅ * | ✅ * |
..contribute more! webrpc generators are just Go templates (similar to Hugo or Helm).
Quick example
Here is an example webrpc schema in RIDL format (a new documentation-like format introduced by webrpc)
webrpc = v1
name = your-app
version = v0.1.0
struct User
- id: uint64
- username: string
- createdAt?: timestamp
struct UsersQueryFilter
- page?: uint32
- name?: string
- location?: string
service ExampleService
- Ping()
- Status() => (status: bool)
- GetUserByID(userID: uint64) => (user: User)
- IsOnline(user: User) => (online: bool)
- ListUsers(q?: UsersQueryFilter) => (page: uint32, users: []User)
error 100 RateLimited "too many requests" HTTP 429
error 101 DatabaseDown "service outage" HTTP 503
Generate webrpc Go server+client code:
webrpc-gen -schema=example.ridl -target=golang -pkg=main -server -client -out=./example.gen.go
and see the generated ./example.gen.go
file of types, server and client in Go. This is essentially
how the golang-basics example was built.
Example apps
Example | Description |
---|---|
hello-webrpc | Go server <=> Javascript webapp |
hello-webrpc-ts | Go server <=> Typescript webapp |
golang-basics | Go server <=> Go client |
golang-nodejs | Go server <=> Node.js (Javascript ES6) client |
node-ts | Node.js server <=> Typescript webapp client |
Why
TLDR; it's much simpler + faster to write and consume a webrpc service than traditional approaches like a REST API or gRPC service.
- Code-generate your client libraries in full -- never write another API client again
- Compatible with the Web. A Webrpc server is just a HTTP/HTTPS server that speaks JSON, and thus all existing browsers, http clients, load balancers, proxies, caches, and tools work out of the box (versus gRPC). cURL "just works".
- Be more productive, write more correct systems.
Writing a Web service / microservice takes a lot of work and time. REST is making me tired. There are many pieces to build -- designing the routes of your service, agreeing on conventions for the routes with your team, the request payloads, the response payloads, writing the actual server logic, routing the methods and requests to the server handlers, implementing the handlers, and then writing a client library for your desired language so it can speak to your Web service. Yikes, it's a lot of work. Want to add an additional field or handler? yea, you have to go through the entire cycle. And what about type-safety across the wire?
webrpc automates a lot the work for you. Now from a single webrpc schema file,
you can use the webrpc-gen
cli to generate source code for:
- Strongly-typed request / response data payloads for your target language
- Strongly-typed server interface and methods on the service, aka the RPC methods
- Complete client library to communicate with the web service
Design / architecture
webrpc services speak JSON, as our goals are to build services that communicate with webapps. We optimize for developer experience, ease of use and productivity when building backends for modern webapps. However, webrpc also works great for service<->service communication, but it won't be as fast as gRPC in that scenario, but I'd be surprised to hear if for the majority of cases that this would be a bottleneck or costly tradeoff.
webrpc is heavily inspired by gRPC and Twirp. It is architecturally the same and has a similar workflow, but simpler. In fact, the webrpc schema is similar in design to protobuf, as in we have messages (structs) and RPC methods, but the type system is arguably more flexible and code-gen tooling is simpler. The webrpc schema is a documentation-like language for describing a server's api interface and the type system within is inspired by Go, Typescript and WASM.
We've been thinking about webrpc's design for years, and were happy to see gRPC and Twirp come onto the scene and pave the way with some great patterns. Over the years and after writing dozens of backends for Javascript-based Webapps and native mobile apps, and even built prior libraries like chi, a HTTP router for Go -- we asked ourselves:
Why have "Rails" and "Django" been such productive frameworks for writing webapps? And the answer we came to is that its productive because the server and client are the same program, running in the same process on the same computer. Rails/Django/others like it, when rendering client-state can just call a function in the same program, the client and the server are within the same domain and same state -- everything is a function-call away. Compare this to modern app development such as writing a React.js SPA or a native iOS mobile app, where the app speaks to an external API server with now the huge added effort to bridge data/runtime from one namespace (the app) to an entirely other namespace (the server). It's too much work and takes too much time, and is too brittle. There is a better way! instead of writing the code.. just generate it. If we generate all of the code to native objects in both app/server, suddenly, we can make a remote service once again feel like calling a method on the same program running on the same computer/process. Remote-Procedure-Call works!
Finally, we'd like to compare generated RPC services (gRPC/Twirp/webrpc/other) to the most common pattern to writing services by "making a RESTful API", where the machinery is similar to RPC services. Picture the flow of data when a client calls out to a server -- from a client runtime proxy-object, we encode that object, send it over the wire, the server decodes it into a server runtime proxy-object, the server handler queries the db, returns a proxy object, encodes it, and sends the function return data over the wire again. That is a ton of work, especially if you have to write it by hand and then maintain robust code in both the client and the server. Ahh, I just want to call a function on my server from my app! Save yourself the work and time, and code-generate it instead - Enter gRPC / Twirp .. and now, webrpc :)
Future goals/work:
- Add RPC streaming support for client/server
- More code generators.. for Rust, Python, ..
Schema
The webrpc schema type system is inspired by Go and TypeScript, and is simple and flexible enough to cover the wide variety of language targets, designed to target RPC communication with Web applications and other Web services.
High-level features:
- RIDL, aka RPC IDL, aka "RPC interface design language", format - a documentation-like schema format for describing a server application.
- JSON schema format is also supported if you prefer to write tools to target webrpc's code-gen tools
- Type system inspired by Go + Typescript
- integers, floats, byte, bool, any, null, date/time
- lists (multi-dimensional arrays supported too)
- maps (with nesting / complex structures)
- structs / objects
- optional fields, default values, and pluggable code-generation for a language target
- enums
For more information please see the schema readme.
Development
Building from source
- Install Go 1.16+
- $
make build
- $
make test
- $
make install
Writing your own code-generator
Authors
- Peter Kieltyka
- José Carlos Nieto
- Vojtech Vitek
- ..and full list of contributors!
Credits
- Twirp authors for making twirp. Much of the webrpc-go library comes from the twirp project.
- gRPC authors, for coming up with the overall architecture and patterns for code-generating the bindings between client and server from a common IDL.
License
MIT