Make it easy to create clean and elegant gRPC based microservices.
This package is written in and for both Node.js and Golang, enabling them to work seamlessly with each other in one codebase.
NOTE: this repo itself is an example of using NgRPC in practice.
First, install the CLI tool via one of the following commands:
npm i -g @ayonli/ngrpc # for Node.js user
go install github.com/ayonli/ngrpc/cli/ngrpc@latest # for Golang user
Then use one of the following commands to initiate the project:
ngrpc init -t node # for Node.js project
ngrpc init -t go # for Golang project
First, take a look at this configuration (ngrpc.json):
{
"$schema": "https://raw.githubusercontent.com/ayonli/ngrpc/main/ngrpc.schema.json",
"protoPaths": [
"proto"
],
"protoOptions": {
"defaults": true
},
"apps": [
{
"name": "example-server",
"url": "grpc://localhost:4000",
"serve": true,
"services": [
"services.ExampleService"
],
"entry": "entry/main.ts", // alternatively, we can use `entry/main.go` instead.
"stdout": "out.log"
},
{
"name": "user-server",
"url": "grpcs://localhost:4001",
"serve": true,
"services": [
"services.UserService"
],
"entry": "entry/main.go",
"stdout": "out.log",
"cert": "certs/cert.pem",
"key": "certs/cert.key",
"ca": "certs/ca.pem"
},
{
"name": "post-server",
"url": "grpcs://localhost:4002",
"serve": true,
"services": [
"services.PostService"
],
"entry": "entry/main.ts",
"stdout": "out.log",
"cert": "certs/cert.pem",
"key": "certs/cert.key",
"ca": "certs/ca.pem"
}
]
}
We have two different entry files here, let's dig in each of them.
import ngrpc from "@ayonli/ngrpc";
if (require.main?.filename === __filename) {
ngrpc.start(ngrpc.getAppName()).then(app => {
process.send?.("ready"); // for PM2 compatibility
app.waitForExit();
}).catch(err => {
console.error(err);
process.exit(1);
});
}
package main
import (
"log"
"github.com/ayonli/ngrpc"
_ "github.com/ayonli/ngrpc/services"
)
func main() {
app, err := ngrpc.Start(ngrpc.GetAppName())
if err != nil {
log.Fatal(err)
} else {
app.WaitForExit()
}
}
-
protoPaths
The directories that stores the.proto
files. Normally,.proto
files are stored in theproto
folder.TIP: don't forget add
--proto_path=${workspaceRoot}/proto
to theprotoc.options
in VSCode's settings in order for the vscode-proto3 plugin to work properly. -
apps
This property configures the apps that this project serves and connects.-
name
The name of the app. -
url
The URL of the gRPC server, supported schemes aregrpc:
,grpcs:
,http:
,https:
orxds:
(in Node.js, make sure package @grpc/grpc-js-xds is installed). -
serve
If this app is served by the NgRPC app server. If this property isfalse
, that means the app is served by other programs and we just connect to it. -
services
The services served by this app. -
entry
The entry file used to spawn apps.-
During development, the entry filename shall be suffixed either by
.ts
or.go
, when running the program, NgRPC automatically compiles the file when needed. -
In production, the entry filename shall the compiled file's name, which is suffixed by
.js
or has no suffix at all (for Golang, or.exe
in Windows).
The program is spawned with the app name in the command line argument, we can use
ngrpc.getAppName()
orngrpc.GetAppName()
to retrieve it. -
-
stdout
Log file used for stdout.
More Options for
apps
-
cert
The certificate filename when using TLS/SSL. -
key
The private key filename when using TLS/SSL. -
ca
The CA filename used to verify the other peer's certificates, when omitted, the system's root CAs will be used.It's recommended that the gRPC application uses a non-public CA, so the client and the server can establish a private connection that no outsiders can join in.
-
stderr
Log file used for stderr. If omitted andstdout
is set, the program usesstdout
forstderr
as well. -
env
Additional environment variables passed to theentry
file.
-
More Top Options
-
namespace
This is the root namespace of the services in Node.js, and the directory that stores the service class (.ts
) files. Normally, this option is omitted and useservices
by default. -
importRoot
Where to begin searching for TypeScript / JavaScript files, the default is.
. If given, there are two rules for setting this option:- During development, if we use a source directory, say
src
(respectively, setcompilerOptions.rootDir
tosrc
intsconfig.json
), then this option should be set tosrc
as well. - In production, if we compile our program to a build directory, say
dist
(respectively, setcompilerOptions.outDir
todist
intsconfig.json
), then this option should be set todist
as well.
This option is also used for generating Golang code from the
.proto
files. If we set asrc
directory, the code will be generated into that directory as well. - During development, if we use a source directory, say
-
protoOptions
These options are used when loading the.proto
files in Node.js. Check ngrpc.schema.json for more details.
In Node.js, services are automatically discoverd and imported when the program starts, in Golang, we
import the services
package and name it _
for its side-effect which registers the services.
Then we use the ngrpc.start()
/ ngrpc.Start()
function to initiate the app with the app name, it
initiates the server (if served) and client connections, prepares the services ready for use.
Next we use the app.waitForExit()
/ app.WaitForExit()
function to wait for the interrupt / exit
signal from the system for a graceful shutdown. In Golang, this function also keeps the program
running and prevent premature exit.
With these simple configurations, we can write our gRPC application straightforwardly in .proto
files and .ts
or .go
files, without any headache of when and how to start the server or
connect to the services, all is properly handled behind the scene.
-
ngrpc init [flags]
initiate a new NgRPC project-
-t --template <string>
available values arego
ornode
TIP: we can run this command twice with different template for the setup for both languages, existing files will be untouched.
-
-
ngrpc start [app]
start an app or all apps (exclude non-served ones)-
app
the app name in the config file
-
-
ngrpc restart [app]
restart an app or all apps (exclude non-served ones)-
app
the app name in the config file
-
-
ngrpc reload [app]
hot-reload an app or all apps-
app
the app name in the config file
NOTE: only Node.js supports hot-reloading, Golang programs just reply that they don't support this feature.
-
-
ngrpc stop [app]
stop an app or all apps-
app
the app name in the config file
-
-
ngrpc list
orngrpc ls
list all apps (exclude non-served ones) -
ngrpc run <filename> [args...]
runs a script file that attaches to the services, can be either Golang (.go
) or Node.js (.ts
) programs. -
ngrpc protoc
generate golang program files from the proto files.NOTE: this command is not used if our project only contains Node.js programs.
-
ngrpc cert <out> [flags]
generate a pair of self-signed certificate.-
--ca string
use a ca.pem for signing, if doesn't exist, it will be auto-generated (defaultcerts/ca.pem
) -
--caKey string
use a ca.key for signing, if doesn't exist, it will be auto-generated (defaultcerts/ca.key
)
-
-
ngrpc host [flags]
start the host server in standalone mode-
--stop
stop the host server
NOTE: when
start
command is issued, the host server will be automatically started. Thehost
command is used when our program isn't started by thestart
command and we need the functionalities that NgRPC provides. For examples, we start our program via PM2 and we still need thereload
command to function once we deployed new updates. -
After we've modified our source code (or recompiled), the .proto
files, or the config file, we can
use the reload
command to hot-reload our apps without restarting the process.
When the command is issued, the application will scan the imported service files and their
dependencies (exclude the ones in node_modules
), and reload them all at once. Since this procedure
doesn't restart the process, all context stores in the global scope are still available.
Hot-reloading is much faster than restarting the whole program, the client will experience
0-downtime of our services.
It's important to point out, though, that the hot-reloading model this package uses only supports services and their dependencies, any other code, for example, the change of the entry file, will not join the reloading process and cannot be hot-reloaded, if such changes are made, a full restart is needed for the new code to run.
Why not auto-reload when the file is changed?
gRPC uses the .proto
file for definition and the .ts
file for implementation, it's hard to keep
track on both files at the same time. If reload immediately after a file is changed, there may be
inconsistency between the two files and causing the program to fail. So this package provides the
reload
command that allows us to manually reload the app when we're done with our changes.
This package uses a host-guest model for process management. When using the start
command to start
the app, the CLI tool also starts a host server to hold communication between apps, the host is
responsible to accept commands sent by the CLI tool and distribute them to the app.
When an app crashes, the host server is also responsible for re-spawning it, this feature guarantees that our app is always online. Except when the host server is running in standalone mode, in which the app should be re-spawned by the external process management like PM2.
Moreover, the CLI tool only works for the app instance, if the process contains other logics
that prevent the process to exit, the stop
command will not be able to terminate the process, in
such case, a force kill is required.
To allow NgRPC to handle the serving and connecting process of our services, we need to implement our services in a well-designed fashion.
For example, a typical service should be designed like this:
import { ServiceClient, service } from "@ayonli/ngrpc";
declare global {
namespace services {
const ExampleService: ServiceClient<ExampleService>;
}
}
@service("services.ExampleService")
export default class ExampleService {
// methods and private fields...
}
If this is a client-side service representation (only for referencing), it should be defined as an abstract class, like this:
import { ServiceClient, service } from "@ayonli/ngrpc";
declare global {
namespace services {
const UserService: ServiceClient<UserService>;
}
}
@service("github.ayonli.ngrpc.services.UserService")
export default abstract class UserService {
// abstract methods...
}
type ExampleService struct {
// A service to be served need to embed the UnimplementedServiceServer.
proto.UnimplementedExampleServiceServer
}
For NgRPC, a client-side service representation struct is needed as well:
// A pure client service is an empty struct, which is only used for referencing to the service.
type ExampleService {}
In each service file, we need to define an init
function to use the service:
func init() {
ngrpc.Use(&ExampleService{})
}
For a service in order to be served, a Serve()
method is required in the service struct:
func (self *ExampleService) Serve(s grpc.ServiceRegistrar) {
proto.RegisterExampleServiceServer(s, self)
// other initiations, e.g. establishing database connections
}
All services (server-side and client-side) must implement the Connect()
method in order to be
connected:
func (self *Service) Connect(cc grpc.ClientConnInterface) proto.ExampleServiceClient {
return proto.NewExampleServiceClient(cc)
}
The service may implement a GetClient()
which can be used to reference the service client
in a more expressive way:
func (self *ExampleService) GetClient(route string) (proto.ExampleServiceClient, error) {
return ngrpc.GetServiceClient(self, route)
}
In Node.js
Simply implement the LifecycleSupportInterface
for the service class, for example:
import { LifecycleSupportInterface, service } from "@ayonli/ngrpc";
@service("services.ExampleService")
export default class ExampleService implements LifecycleSupportInterface {
async init(): Promise<void> {
// When the service is loaded (or reloaded), the `init()` method will be automatically
// called, we can add some async logic inside it, for example, establishing database
// connection, which is normally impossible in the default `constructor()` method
// since it doesn't support asynchronous codes.
}
async destroy(): Promise<void> {
// When the app is about to stop, or the service is about to be reloaded, the `destroy()`
// method will be called, which gives the ability to clean up and release resources.
}
}
In Golang
We use the Serve()
method for additional setup, and the Stop()
method for teardown.
func (self *ExampleService) Serve(s grpc.ServiceRegistrar) {
proto.RegisterExampleServiceServer(s, self)
// other initiations, e.g. establishing database connections
}
func (self *ExampleService) Stop() {
// release database connections, etc.
}
In Node.js
Just add a private property in the class that points to another service, like this:
@service("github.ayonli.ngrpc.services.PostService")
export default class PostService {
private userSrv = services.UserService;
// other private properties...
async getPost(query: PostQuery): Promise<Post> {
const post = this.postStore?.find(item => item.id === query.id);
if (post) {
// ---- highlight ----
const author = await this.userSrv.getUser({ id: post.author });
// ---- highlight ----
return { ...post, author };
} else {
throw new Error(`Post ${query.id} not found`);
}
}
}
In Golang
Just add an exported field that points to another service, like this:
type UserService struct {
proto.UnimplementedUserServiceServer
PostSrv *PostService // set as exported field for dependency injection
// other non-exported fields...
}
func (self *UserService) GetMyPosts(ctx context.Context, query *proto.UserQuery) (*proto.PostQueryResult, error) {
return goext.Try(func() *services_proto.PostQueryResult {
user := goext.Ok(self.GetUser(ctx, query))
// ---- highlight ----
ins := goext.Ok(self.PostSrv.GetClient(user.Id))
res := goext.Ok(ins.SearchPosts(ctx, &services_proto.PostsQuery{Author: &user.Id}))
// ---- highlight ----
return (*services_proto.PostQueryResult)(res)
})
}
NgRPC comes with a client-side load balancer, we can set the same service in multiple apps and the internal routing resolver which automatically redirect traffic for us.
There are three algorithms used based on the route
:
- When
route
is not empty:- If it matches one of the name or URL of the apps, the traffic is routed to that app directly.
- Otherwise the program hashes the route string against the apps and match one by the mod value
of
hash % active_nodes
.
- When
route
is empty, the program uses round-robin algorithm against the active nodes.
In Node.js
To use this feature, define the request message that extends / augments the interface
RoutableMessageStruct
, it contains a route
key that can be used in the internal client load
balancer. When the client sending a request which implements this interface, the program will
automatically route the traffic to a certain server evaluated by the route
key.
In Golang
To use this feature, we need to provide the route
argument when calling the
ngrpc.GetServiceClient()
function or the GetClient()
of the service struct.
For Example:
The .proto
File
message RequestMessage = {
string route = 1;
// other fields
};
In Node.js
// the .ts file
import { RoutableMessageStruct } from "@ayonli/ngrpc";
export interface RequestMessage extends RoutableMessageStruct {
// other fields
}
In Golang
func main() {
msg := &proto.RequestMessage{
Route: "route key"
// other fields
}
ins := ngrpc.GetServiceClient(&services.SomeService{}, msg.Route)
}
Apart from the client-side load balancing, server-side load balancing is automatically supported by
gRPC, either by reverse proxy like NGINX or using the xds:
protocol for Envoy Proxy.
It it possible to start an app without providing the name, such an app will not start the server, but only connects to the services. This is useful when we're using gRPC services in a frontend server, for example, a web server, which only handles client requests and direct calls to the backend gRPC services, we need to establish connection between the web server and the RPC servers, but we don't won't to serve any service in the web server.
The following apps do not serve, but connect to all the services according to the configuration file. We can do all the stuffs provided by NgRPC in the web server as we would in the RPC server, because all the differences between the gRPC client and the gRPC server are hidden behind the scene.
In Node.js
import ngrpc from "@ayonli/ngrpc";
(async () => {
const app = await ngrpc.start();
})()
In Golang
import "github.com/ayonli/ngrpc"
func main() {
app, err := ngrpc.Start("")
}
Apart from the unnamed app, an app can be configured with serve: true
but no services, such an app
does not actually start the gRPC server neither consume the port. But such an app can be used, say,
to start a web server, which connects to the gRPC services and uses the facility this package
provides, such as the CLI tool and the hot-reloading model.
For example:
// ngrpc.json
{
// ...
"apps": [
{
"name": "web-server",
"url": "http://localhost:3000",
"serve": true,
"services": [], // leave this blank
// ...
},
// ...
]
}
In Node.js
Check out web/main.ts.
In Golang
Check out web/main.go.
Sometimes it's just useful that we can write a simple script program that connects to the services and call their methods, for whatever the reason is, NgRPC makes this very easy for us.
In Node.js
import ngrpc from "@ayonli/ngrpc";
ngrpc.runSnippet(async () => {
const reply = await services.ExampleService.sayHello({ name: "World" });
console.log(reply.message);
});
In Golang
package main
import (
"context"
"fmt"
"github.com/ayonli/goext"
"github.com/ayonli/ngrpc"
"github.com/ayonli/ngrpc-test/services"
"github.com/ayonli/ngrpc-test/services/proto"
)
func main() {
done := ngrpc.ForSnippet()
defer done()
ctx := context.Background()
expSrv := goext.Ok((&services.ExampleService{}).GetClient(""))
reply := goext.Ok(expSrv.SayHello(ctx, &proto.HelloRequest{Name: "World"}))
fmt.Println(reply.Message)
}
In production, instead of using the built-in process management feature, we can use PM2 to govern our program, which provides more features such as monitoring, log-rotation, web admin, etc. This package is designed to work well with PM2, and not only Node.js, but the Golang program can be hosted by PM2, too.
To enable working with PM2, use the ngrpc.loadConfigForPM2()
function in the ecosystem.config.js
(or a custom filename), it loads the configurations and reorganizes them so that the same
configurations can be used in PM2's configuration file. Moreover, it has done some work internally
to support Golang programs for PM2, regardless of whether it has been compiled or not.
For example:
// ecosystem.config.js
const { default: ngrpc } = require("@ayonli/ngrpc");
module.exports = ngrpc.loadConfigForPM2();
In order to code a clean and elegant gRPC based application, apart from the features that NgRPC provides, we can order our project by performing the following steps.
-
Uses the
proto
folder to store all the.proto
files in one place (by default). -
Uses the
services
folder for all the service files (by default), the namespace / package of those files should be the same as the folder's name (which is alsoservices
).NOTE: although sub-folders and sub-namespaces / sub-packages are supported, it's a little tricky in Golang, try to prevent this as we can.
-
Design the
.proto
files with a reasonable scoped package name, don't just name itservices
, instead, name it something like[org].[repo].services
,.proto
files should be shared and reused across different projects, using a long name to prevent collision and provide useful information about the service. Respectively, the directory path should reflect the package name. See the proto files of this project as examples. -
In Node.js, use the same file structures and symbol names (as possible as we can) in the class files to reflect the ones in the
.proto
files, create a consistent development experience.In Golang, Always implement the
GetClient()
method in the service and use an exported field in the service struct to reference to each other (for dependency injection).
For Node.js, see api-node.md.
For Golang, see the package detail.