Awesome
ngx-grpc
Angular gRPC framework.
Features
- two-way-binding-friendly protobuf messages implementation (instead of Java-like setters / getters / builder pattern in original google-protobuf)
- client services are wired to Angular's dependency injection
- typescript first-class support
- rxjs first-class support
- client & bidirectional streaming (only possible with @improbable-eng/grpc-web)
- interceptors
- logger
- support for well-known types, including
Any
- support for JSON mappings
- 3 different client implementations: grpc-web, @improbable-eng/grpc-web and web worker
- easy to install, update and support thanks to npm packages
Example
The example requires docker & docker-compose to be installed
Clone this repository and run
npm ci
npm run build
npm run examples
On m1 chips replace npm ci
with npm ci --target_arch=x64 --no-optional
.
Now open your browser at http://localhost:4200/. The source code could be found at examples directory.
Installation
The documentation uses @ngx-grpc/grpc-web-client by default, however is applicable to any client you choose.
First ensure that you
- installed
protoc
as a binary or as a part ofgrpc-tools
npm package - configured your backend / grpc-web proxy according to grpc-web documentation or follow documentation of the alternative client if you use one
Then in your Angular project's root directory run
npm i -S @ngx-grpc/common @ngx-grpc/core @ngx-grpc/grpc-web-client @ngx-grpc/well-known-types google-protobuf grpc-web
npm i -D @ngx-grpc/protoc-gen-ng @types/google-protobuf
Where:
- @ngx-grpc/common contains common reusable types for other ngx-grpc packages
- @ngx-grpc/core contains angular specific implementation
- @ngx-grpc/grpc-web-client provides a client based on grpc-web
- @ngx-grpc/protoc-gen-ng generates the code based on your proto files
- @ngx-grpc/well-known-types contains well-known types
- google-protobuf is required to encode / decode the messages
- grpc-web implements the transport between the browser and grpc proxy
Also you can choose between alternative client implementations:
- @ngx-grpc/grpc-web-client - based on grpc-web
- @ngx-grpc/improbable-eng-grpc-web-client - alternative client implementation based on @improbable-eng/grpc-web
- @ngx-grpc/worker-client - similar to @ngx-grpc/grpc-web-client but running in worker
Generate the code
MacOS / Linux
Add proto:generate
script to your package.json
scripts
section:
{
"scripts": {
"proto:generate": "protoc --plugin=protoc-gen-ng=$(which protoc-gen-ng) --ng_out=<OUTPUT_PATH> -I <PROTO_DIR_PATH> <PROTO_FILES>"
}
}
Where:
OUTPUT_PATH
- the directory your code will be generated at (please ensure the directory exists, otherwise you'll have aprotoc
error)PROTO_DIR_PATH
- the root path of your proto filesPROTO_FILES
- list of proto files to use
Example:
{
"scripts": {
"proto:generate": "protoc --plugin=protoc-gen-ng=$(which protoc-gen-ng) --ng_out=./src/proto -I ../proto $(find ../proto -iname \"*.proto\")"
}
}
Finally, run npm run proto:generate
every time you want to (re)generate the code
Advanced generator config
You can have more control on what and how is being generated. Create a ngx-grpc.conf.js
(.json
format also supported) file in your project root.
E.g. to generate well-known types in your project instead of using @ngx-grpc/well-known-types
, use this config:
module.exports = {
embedWellKnownTypes: true,
};
More details on the configuration properties and their default values see here.
Then update your package.json command with path to this file config=./ngx-grpc.conf.js
:
{
"scripts": {
"proto:generate": "protoc --plugin=protoc-gen-ng=$(which protoc-gen-ng) --ng_out=config=./ngx-grpc.conf.js:./src/proto -I ../proto $(find ../proto -iname \"*.proto\")"
}
}
Windows
Unfortunately the way to generate files on Windows slightly differs. Here is a sophisticated example that shows how to scan windows folder with proto files and pass it to protoc-gen-ng.
{
"scripts": {
"proto:generate:win": "for /f %G in ('dir /b ..\\proto\\*.proto') do grpc_tools_node_protoc --plugin=protoc-gen-ng=.\\node_modules\\.bin\\protoc-gen-ng.cmd --ng_out=.\\output\\path -I ..\\proto ..\\proto\\%G",
}
}
Usage
Import the required modules
In general, you need to import at least
- GrpcCoreModule
- GrpcWebClientModule (or another client implementation, see below)
@NgModule({
imports: [
GrpcCoreModule.forRoot(),
GrpcWebClientModule.forRoot({
settings: { host: 'http://localhost:8080' },
}),
],
})
export class AppModule {}
You also can define them in child modules by using forChild()
methods instead of forRoot()
.
Per-service clients configuration
Instead of configuring the client settings globally you can configure them per-service. Every service has an injected configuration which could be found e.g. in the corresponding *.pbconf.ts
file.
E.g. for a service TestServiceClient
you need to provide the GRPC_TEST_SERVICE_CLIENT_SETTINGS
:
@NgModule({
providers: [
{ provide: GRPC_TEST_SERVICE_CLIENT_SETTINGS, useValue: { host: 'http://localhost:8080' } as GrpcWebClientSettings },
],
})
export class AppModule {}
To set grpcweb / binary proto format use
{ provide: GRPC_TEST_SERVICE_CLIENT_SETTINGS, useValue: { host: 'http://localhost:8080', format: 'binary' } as GrpcWebClientSettings },
From now on this particular service is set.
Service client methods
Concept
Every client call accepts a message to be sent and returns an Observable
of response message(s). However, the request is not being executed until the Observable
gets subscribed, so it can be safely used at any place.
To cancel the request / close the connection simply unsubscribe from that Subscription
. Of course, if connection is closed by the server or the error happens, the Observable
gets terminated.
class MyService {
constructor(private client: EchoClient) {}
sendOne() {
this.client.echo(new EchoRequest({ message: 'text' })).subscribe(res => console.log(res));
// or if you want to terminate it, e.g. it is a server stream or you navigate away and do not need to wait
const sub = this.client.echo(new EchoRequest({ message: 'text' })).subscribe(res => console.log(res));
setTimeout(() => sub.unsubscribe(), 1000); // this closes connection
}
}
The behaviour above is possible due to the Observable
's natural laziness and ability to be terminated.
The server streaming has the same signature as the unary requests, but the returned Observable
can emit more than one message and the connection is kept open.
The client streaming is differrent, because it accepts an Observable
(and any of its derivatives, such as Subject
, BehaviourSubject
, etc.) of messages
class MyService {
constructor(private client: EchoClient) {}
sendMany() {
const stream = from(['message 1', 'message 2', 'message 3']);
this.client.echoMany(stream).subscribe(res => console.log(res));
}
}
The bidirectional streaming has the same signature as the client's one and is a combination of server and client streaming.
Implementation details
Each RPC has two corresponding methods.
- the first, that emits messages, is a direct method of the service client.
- the second, that emits events, is scoped into service client property
$raw
.
E.g. for rpc Echo(...)
there would be the following:
echo(...)
- returnsObservable
of messages and throws errors in case of non-zero status codes. This is the most common use-case.$raw.echo(...)
- returnsObservable
ofGrpcEvent
s. Events could be of two kinds:GrpcDataEvent
containing the message inside andGrpcStatusEvent
containing gRPC status response. Apart from the returned data type there is important difference in the behaviour. There are no errors thrown in this stream (by design). All errors are considered to be normalGrpcStatusEvent
s. Furthermore, this method is the only one where it is anyhow possible to read the gRPC status code0
(OK
) metadata. This method is not that comfortable to use in every place, but it can do things that are not achievable with the method above.
There are two custom RxJS operators that could be used on the stream to make it easier:
throwStatusErrors
- searches for the non-zero status codes and throws them as errorstakeMessages
- searches for the messages
For usage example look at any of your generated .pbsc.ts
file. In fact, those two operators turn the raw method into a 'normal' one.
Messages
To create a new message just pass its initial values to the constructor: new Message(myInitialValues)
. Here is some information on the message's methods:
constructor
- accepts a message or an object with initial values. All values are safely / deeply cloned.toObject()
- casts message as is to a normal JavaScript objecttoJSON()
- convenience method to be able to pass message toJSON.stringify(msg)
toProtobufJSON()
- constructs a protobuf-defined JSON. Accepts an optionalGrpcMessagePool
(seegoogle.protobuf.Any
section) which is required only if the message or some of its descendants embedsgoogle.protobuf.Any
Int64
JavaScript does not support int64
, out of the box, that's why all of its kinds are generated as string by default.
You can however override this behavior by passing JS_NUMBER
or JS_STRING
option to the appropriate field. Example:
message Message {
int64 bigInt = 1 [jstype = JS_NUMBER];
uint64 bigUint = 2 [jstype = JS_NUMBER];
}
Well-known types
The well-known types are served as a separate package. You can also configure generation of the well-known types together with your proto definitions (like older versions did).
Some types have additional functionality, see below.
google.protobuf.Any
The google.protobuf.Any
has additional methods pack
and unpack
.
Unpacking the message requires a special message pool GrpcMessagePool
where the expected message types are listed; otherwise the unpacking would not be possible.
Example of type-safe unpacking:
// we expect one of 3 message types to be packed into Any
const myAny: Any;
const pool = new GrpcMessagePool([Empty, Timestamp, MyMessage]);
try {
switch(myAny.getPackedMessageType(pool)) {
case Empty: console.log('Empty found', myAny.unpack<Empty>(pool)); break;
case Timestamp: console.log('Timestamp found', myAny.unpack<Timestamp>(pool)); break;
case MyMessage: console.log('MyMessage found', myAny.unpack<MyMessage>(pool)); break;
default: console.log('No packed message inside');
}
} catch (ex) {
console.error('Something went wrong, e.g. packed message definition is not in the pool');
}
google.protobuf.Timestamp
The google.protobuf.Timestamp
has additional methods to cast from / to Date
and ISO string date representation.
Custom well-known types
For well-known types that are not part of the google.protobuf
package, you can override the imports to use for specific packages.
This is especially useful if you are using a protobuf schema registry like Buf for sharing some common messages through different projects.
Example of a custom well-known type configuration:
module.exports = {
customWellKnownTypes: {
"company.internal.commons": "@company-internal/grpc-commons"
}
}
This will change all import statements that reference a message of the package company.internal.commons
to use @company-internal/grpc-commons
, instead of the relative file path.
If the embedWellKnownTypes
configuration is enabled, the customWellKnownTypes
configuration will be ignored and the messages will be generated as usual.
Interceptors
You can add global interceptors to all gRPC calls like Angular's built-in HttpClient
interceptors.
The important differences
- unlike
HttpClient
interceptorsGrpcInterceptor
s need to work with event streams; there are no errors thrown. Instead you should listen to theGrpcStatusEvent
and decide whether it is an error or not. Please keep this in mind - the incoming data can be a message or a stream of messages (in case of client streaming)
As an example see GrpcLoggerInterceptor
in the core package.
Logger
You can enable logging using GrpcLoggerInterceptor
(provided by @ngx-grpc/core). Add to your AppModule
the following import:
GrpcLoggerModule.forRoot(),
Then open the browser console and you should see all the requests and responses in a readable format.
Optionally, you can provide a more detailed configuration. Example:
GrpcLoggerModule.forRoot({
settings: {
// enables logger in dev mode and still lets you see them in production when running `localStorage.setItem('logger', 'true') in the console`
enabled: localStorage.getItem('logger') === 'true' || !environment.production,
// protobuf json is more human-readable than the default toObject() mapping
// please beware: if you use google.protobuf.Any you must pass the proper `messagePool` argument
requestMapper: (msg: GrpcMessage) => msg.toProtobufJSON(),
responseMapper: (msg: GrpcMessage) => msg.toProtobufJSON(),
},
}),
Alternative client: @improbable-eng/grpc-web
The alternative grpc-web implementation from Improbable Engineering provides way more features than standard grpc-web from Google. It supports various transports including WebSocket-based and even Node (can be useful e.g. for SSR).
The only client that supports client / bidirectional streaming. This however also requires the server to be able to handle websocket transport. For this purpose improbable-eng team introduced grpc-web-proxy - a special facade for the normal grpc server that acts like envoy, but has also the ability to handle websocket transport.
Installation:
npm i -S @ngx-grpc/improbable-eng-grpc-web-client @improbable-eng/grpc-web
Then configuration is similar to the other clients, however there is a transport to configure:
import { grpc } from '@improbable-eng/grpc-web';
import { GrpcCoreModule } from '@ngx-grpc/core';
import { ImprobableEngGrpcWebClientModule } from '@ngx-grpc/improbable-eng-grpc-web-client';
const xhr = grpc.CrossBrowserHttpTransport({});
const ws = grpc.WebsocketTransport();
@NgModule({
imports: [
GrpcCoreModule.forRoot(),
ImprobableEngGrpcWebClientModule.forChild({
settings: {
host: 'http://localhost:8080',
// we might want to use different transports as recommended by improbable-eng team
// because websocket transport acts a bit differently and is intended for client streaming only
transport: {
unary: xhr,
serverStream: xhr,
clientStream: ws,
bidiStream: ws,
},
// or simply e.g.
// transport: ws, // to configure all methods to use websockets
},
}),
],
})
export class AppModule {}
Web worker
Web worker allows to run gRPC clients, messages serialization and deserialization in a separate thread. It might give some performance benefits on large data sets; however the main reason of the worker is to avoid blocking the main thread. That means that rendering engine has more resources to work on rendering while the messages processing is done in parallel.
First, install additional packages:
npm i -S @ngx-grpc/worker @ngx-grpc/worker-client
Then configure the web worker. First you need to adapt the code generation settings (see above) to generate pbwsc
files. These files will contain the worker service client definitions.
Now, generate the worker (angular cli), e.g. with the name grpc
:
ng g web-worker grpc
or for Angular < 9
ng g worker grpc
You should see grpc.worker.ts
close to your app.module.ts
. Open this file and replace the contents with the following:
/// <reference lib="webworker" />
import { GrpcWorker } from '@ngx-grpc/worker';
import { GrpcWorkerEchoServiceClientDef } from '../proto/echo.pbwsc';
const worker = new GrpcWorker();
worker.register(
// register here all the service clients definitions
GrpcWorkerEchoServiceClientDef,
);
worker.start();
Finally use the following imports:
@NgModule({
imports: [
GrpcCoreModule.forRoot(),
GrpcWorkerClientModule.forRoot({
worker: new Worker('./grpc.worker', { type: 'module' }),
settings: { host: 'http://localhost:8080' },
}),
],
})
export class AppModule {
}
That's it. All your requests are served by worker.
Not implemented (yet)
Contributing
- to run tests on Apple m1 chips use
npm ci --no-optional
andbrew install protoc-gen-grpc-web