As we know, each Node.js process runs in a single thread. Usually, we split a single process into multiple processes to take advantage of multi-core systems. On the other hand, it brings more system overhead, sush as maintaining more TCP connections between servers.
This module is designed to share connections among multi-process Nodejs.
Allow ONLY one process "the Leader" to communicate with server. Other processes "the Followers" act as "Proxy" client, and forward all requests to Leader.
The Leader is selected by "Port Competition". Every process try to listen on a certain port (for example 7777), but ONLY one can occupy the port, then it becomes the Leader, the others become Followers.
TCP socket connections are maintained between Leader and Followers. And I design a simple communication protocol to exchange data between them.
If old Leader dies, one of processes will be selected as the new Leader.
'use strict';const co = require('co');const Base = require('sdk-base');const cluster = require('cluster-client');/** * Client Example */class YourClient extends Base { constructor(options) { super(options); this.options = options; this.ready(true); } subscribe(reg, listener) { // subscribe logic } publish(reg) { // publish logic } * getData(id) { // invoke api } getDataCallback(id, cb) { // ... } getDataPromise(id) { // ... }}// create some client instances, but only one instance will connect to serverconst client_1 = cluster(YourClient) .delegate('getData') .delegate('getDataCallback') .delegate('getDataPromise') .create({ foo: 'bar' });const client_2 = cluster(YourClient) .delegate('getData') .delegate('getDataCallback') .delegate('getDataPromise') .create({ foo: 'bar' });const client_3 = cluster(YourClient) .delegate('getData') .delegate('getDataCallback') .delegate('getDataPromise') .create({ foo: 'bar' });// subscribe informationclient_1.subscribe('some thing', result => console.log(result));client_2.subscribe('some thing', result => console.log(result));client_3.subscribe('some thing', result => console.log(result));// publish dataclient_2.publish('some data');// invoke methodclient_3.getDataCallback('some thing', (err, val) => console.log(val));client_2.getDataPromise('some thing').then(val => console.log(val));co(function*() { const ret = yield client_1.getData('some thing'); console.log(ret);}).catch(err => console.error(err));
API
delegate(from, to):
create delegate method, from is the method name your want to create, and to have 6 possible values: [ subscribe, unSubscribe, publish, invoke, invokeOneway, close ], and the default value is invoke
override(name, value):
override one property
create(…)
create the client instance
close(client)
close the client
APIClientBase a base class to help you create your api client
Best Practice
DataClient
Only provider data API, interact with server and maintain persistent connections etc.
No need to concern cluster issue
APIClient
Using cluster-client to wrap DataClient
Put your bussiness logic here
DataClient
const Base = require('sdk-base');class DataClient extends Base { constructor(options) { super(options); this.ready(true); } subscribe(info, listener) { // subscribe data from server } publish(info) { // publish data to server } * getData(id) { // asynchronous API }}
APIClient
const DataClient = require('./your-data-client');const { APIClientBase } = require('cluster-client');class APIClient extends APIClientBase { constructor(options) { super(options); this._cache = new Map(); } get DataClient() { return DataClient; } get delegates() { return { getData: 'invoke', }; } get clusterOptions() { return { name: 'MyClient', }; } subscribe(...args) { return this._client.subscribe(...args); } publish(...args) { return this._client.publish(...args); } * getData(id) { // write your business logic & use data client API if (this._cache.has(id)) { return this._cache.get(id); } const data = yield this._client.getData(id); this._cache.set(id, data); return datal }}