Node.js can be a bit like the wild west when it comes to organising your application structure. Although 'require' goes a long way toward helping you have a good layout, it is not particularly useful when it comes to questions of application lifecycle. That's where this project comes in.
Node's asynchronous nature can make it difficult to be certain that all of your resources are started and running before you start using them. Using lifecycle, you can specify how your application elements depend on each other and you can be guaranteed that any dependencies will be initialised and started in the right order. For example, if module A depends on modules B and C, and C depends on D, then the system will initialise D first, then C and B, then finally A.
Using it is a simple matter of having an object that inherits from the Lifecycle class, and, if you want, overriding any of the lifecycle methods.
var Lifecycle = require("ecosystem").Lifecycle;
var util = require("util");
// You need a constructor that calls the Lifecycle constructor
function MyService(name) {
Lifecycle.call(this, name);
}
// ... and you need to inherit from Lifecycle
util.inherits(MyService, Lifecycle);
...
// ... and if you export your class as "Lifecycle", ecosystem
// will know how to init/start/stop it easily
module.exports = {
Lifecycle: MyService,
...
};There are four methods that your modules can override to implement their lifecycle behaviours:
-
dependencies() - return an array of names of modules that this module depends on. The dependencies will be initialised before this module.
-
init(config, modules, next) - this will be invoked if present on your module, allowing your component to initialise itself from config values, set up various parameters, and so on. This will only be called once. You must call the next() function when you are done, or the init call-chain will terminate.
-
start(next) - here you can actually start your service, connect to a database, set up timers, etc. When you are done, you must invoke the next() function in order for the start call-chain to continue.
-
stop(next) - This will be called when the service is being stopped. Note that a service will only be stopped after all its dependencies are stopped. You must call next() or the stop call-chain will terminate.
Build a dictionary of name-to-component mappings:
var modules = {
foo: InstanceOfFoo,
bar: InstanceOfBar,
...
};Initialise them all this way:
ecosystem.initAll(config, modules, function() {
// this will be called when all modules have been initialised
});Start them all
ecosystem.startAll(modules, function() {
// this will be called when all modules
// have been successfully started
});Stop them when you're done
ecosystem.stopAll(modules, function() {
// This will be called when all
// modules have been stopped.
});So all together:
ecosystem.initAll(config, modules, function() {
// this will be called when all modules have been initialised
ecosystem.startAll(modules, function() {
// this will be called when all modules
// have been successfully started
ecosystem.stopAll(modules, function() {
// This will be called when all
// modules have been stopped.
});
});
});Note that the config should only be passed into the initAll() function; modules can take what they need from the config during initialisation.
Here's an example MySQL service that other modules can declare as a dependency. Then all dependent modules can be assured that when their start() method is called, this service's start() method has already been called. You can find this example in the examples/mysql directory.
var mysql = require("mysql");
var util = require("util");
var _ = require("underscore");
var Lifecycle = require("ecosystem").Lifecycle;
// Construct the service
function MySQL(name) {
Lifecycle.call(this, name);
this._connection = null;
}
util.inherits(MySQL, Lifecycle);
// Your service can optionally implement any of these
// lifecycle methods:
//
// * init(config, modules, next)
// * start(next)
// * stop(next)
_.extend(MySQL.prototype, {
// init is guaranteed to be called before the init() of any
// modules that depend on this module.
init: function(config, modules, next) {
this._connection = mysql.createConnection(config.mysql);
next();
},
// start is guaranteed to be called before the start() of
// any modules that depend on this module; this means that
// a dependent module can assume that the MySQL connection
// will be alive and can use it within the start() method.
start: function(next) {
var that = this;
this._connection.connect(function(err) {
if (err) {
console.error(err);
throw new Error(err);
}
console.log("MySQL connected.");
next();
});
},
stop: function(next) {
this._connection.end(function(err) {
console.log("MySQL connection closed.");
next();
});
},
... your own methods here to talk to the service,
... run queries, etc. etc.
});
module.exports = {
Lifecycle: MySQL
};Now you can use this service from your other code:
var _ = require("underscore");
var util = require("util");
var Lifecycle = require("ecosystem").Lifecycle;
function FooService(name) {
Lifecycle.call(this, name);
...
}
util.inherits(FooService, Lifecycle);
_.extend(FooService.prototype, {
dependencies: function() {
// if you use loadAll() to load dependencies, you declare
// them using the path that they were "require"d with.
// Otherwise, you just use the name that they were constructed
// with.
return [ "./mysql", "./something-else", ];
},
init: function(config, modules, next) {
console.log("Foo service initialised");
next();
},
start: function(next) {
// Here we are guaranteed to have a live mysql
// connection (assuming nothing went wrong).
var mysql = this.dependency("mysql");
mysql.doSomeStuff(function() {
next();
});
},
... More FooService stuff here
});
module.exports = {
Lifecycle: FooService
};In your FooService's start() method, you can see that it can access its "mysql" dependency by name, and then call methods on it. The MySQL service's start() method will have already been called and if all went well, the connection will be live and ready.
There are many ways to start all your services, but here's one easy way to do it:
// app.js
var ecosystem = require("ecosystem");
// Presumably you'd pull this in from a file or something:
var config = {
mysql: {
host: "localhost",
user: "sneelock",
password: ...
database: "circus"
},
...
};
var moduleNames = [
'./mysql',
'./foo',
];
// This convenience will "require" all the listed
// modules and create uninitialised instances of
// all of them. You don't have to use this but it
// saves a lot of boilerplate.
var modules = ecosystem.loadAll(moduleNames);
// This is a standard pattern to initialise your
// modules first, then once that succeeds, you
// can start them all.
ecosystem.initAll(config, modules, function() {
ecosystem.startAll(modules, function() {
console.log("All modules started.");
});
});
// Trap termination signals and close gracefully
function terminate() {
console.log('Shutting down...');
ecosystem.stopAll(modules);
setTimeout(process.exit, 2000);
}
process.on('SIGINT', terminate);
process.on('SIGTERM', terminate);- Write some decent documentation
This is loosely based on the awesome Clojure project, component, by Stuart Sierra.