cluster-mode

©Nobuyori Takahashi < [email protected] >

A cluster process management module for node.js application.

This module's intention is not to replace the built-in cluster module, but to extend it for usability and functionality.

What Cluster-Mode Does

• Manages starting and graceful shutdown of node.js cluster including optional auto respawning of dead worker processes

• Extends cluster process communication including request/response type communication and relaying messages from worker to another worker

• Worker role management: Each worker can assign themselves to be a specific roled worker process

How To Install

npm install cluster-mode

How To Use

var cluster = require('cluster-mode');
var config = {
	max: 8 // start the application with 8 workers
	logger: require('bunyan').createLogger({ name: 'myClusterApp' })
};
cluster.start(config, function () {
	// cluster process is ready
});

Methods

Avialable functions of this module.

.addShutdownTask(task [Function], runOnMaster [*Boolean])

Adds a function to be executed before your application process exits.

Task function will recieve 1 argument, a callback function.

It is useful when you need to clean up your application before terminating the prrocess.

If runOnMaster is set to false, the shutdown task function will NOT be executed on master process.

The default value of runOnMaster is true.

NOTE: The function returns a boolean. When the function succeeds in adding a new shutdown task function, it returns true.

Exmaple:

var cluster = require('cluster-mode');
var success = cluster.addShutdownTask(function (cb) {
	// do what needs to be done
	// when all is finished, move on to next
	cb();
}, false);

// success would be true, if it added the task function
// false means that is task function will NOT be executed on master process

.onExit(task [Function])

Assigns a function to be executed at the moment of exitting of the application process.

The difference from .addShutdownTask() is that .onExit() will be executed after all shutdown tasks.

.start(config [*Object], callback [*Function])

You must invoke this function in order to start your application process.

If you do not pass config, cluster-mode will fall back to its default settings:

NOTE: The optional callback is called when the process is ready.

Default

{
	max: <number of CPU available>,
	log: null,
	autoSpawn: false,
	sync: true
}

Config Object

{
	max: <number> // max number of worker processes to spawn
	log: <object> // logging module object. cluster-mode supports bunyan, winston, log4js, and gracelog
	autoSpawn: <boolean> // automatically re-spawn dead worker processes
	sync: <boolean> // synchronize worker map: default is true
}

max (Required)

This property manages the number of worker processes you want to start with.

NOTE: Set this value to 0 to start your application in non-cluster mode (no workers).

logger (Optional)

cluster-mode module supports, bunyan, winston, log4js and gracelog.

autoSpawn (Optional)

If set to true, cluster-mode will automatically re-spawn a new worker to take place of the dead worker.

If workers die in less than 10 seconds, however, it will consider, there is something wrong with the application and will NOT re-spawn a new worker.

sync (Optional)

A boolean flag to turn on/off auto-synchronization of worker map among all worker processes.

The default is true.

.stop(error [*Error Object])

If invoked from cluster worker, it will send a request to terminate the process to master.

NOTE: If an error object is passed, cluster-module will log an error log and terminates the process with FATAL ERROR.

.terminate()

Terminates the worker process that it calls. This method is only available in cluster mode worker processes.

.isMaster()

Returns true if your process is in cluster-mode (with workers) and a master process.

.isCluster()

Returns true if your process is running with workers.

.getWorkers()

If sync option in configuration object for .start() is true, It will return the worker map, but if it is false, it will return an empty object.

Returns a map of all available workers.

The keys of the map are worker IDs.

NOTE: The map is synchronized from master process asynchronously.

.send(workerId [number], message [object])

Sends a message object to a specific worker process.

The sent message can be caught by message event in the targeted worker process.

.getWorkerIdByPid(pid [number])

Returns the pid of a worker by the pid(process ID) given.

If invalid pid is given, the function will return null.

NOTE: If sync of the configuration is false, the function returns null.

.getWorkerPidById(workerId [number])

Returns the pid (process ID) of a worker by the ID (worker ID) given.

If invalid ID is given, the function will return null.

NOTE: If sync of the configuration is false, the function returns null.

.registerCommand(cmd [String], handler [Function])

Registers a command handler in master process.

The handler will have message object and callback function passed.

NOTE: If you register a command in worker, .sendCommandToWorker() must be called from master process.

var cluster = require('cluster-mode');
cluster.config({
	max: 4
});

if (cluster.isMaster()) {
	cluster.registerCommand('addSharedData', function (msg, callback) {
		sharedData[msg.id] = msg.value;
		callback(null, { result: 'OK' });
	});
}

if (!cluster.isMaster()) {
	cluster.sendCommand('addSharedData', { id: 'xxx', value: 100 }, function (error, res) {
		if (error) {
			// oh no..
		}
		if (res.result === 'OK') {
			// yay
		}
	});
}

.sendCommand(cmd [String], data [Object], callback [Function]);

Requests the master process to execute pre-registered command handler from a worker.

NOTE: This is only available in worker process ONLY.

cluster.sendCommand('getSharedData', { id: 'xxx' }, function (error, res) {
	if (error) {
		// not good
	}
	// yay
	console.log(res.value);
});

.sendCommandToWorker(worker [Object], cmd [String], data [Object], callback [Function]);

Requests the worker process to execute pre-registered command handler and respond back.

NOTE: This is only available in master process ONLY.

.id()

Returns its own worker ID

It returns null for master process.

.workers()

Returns a map of worker objects.

NOTE: Only available in master process.

.registerRole(roleName [String/Array], callback [Function])

Registers a role. It means that the worker process will be refered to as the given roleName.

The callback will have an error if the role is already taken or failed to register.

If given an array of role names, it will try to register the first available role and return the assigned role name.

NOTE 1: This is available in worker process ONLY

NOTE 2: A role is unique to a woker. You may NOT have duplicated roles or share the same role with other workers.

Example with string:

cluster.registerRole('HERO', function (error, registedRole) {
	if (error) {
		// oh no...
	}
	// HERO
	console.log('I am', registeredRole);
});

Example with array:

cluster.registerRole(['HERO', 'VILLAIN'], function (error, registedRole) {
	if (error) {
		// oh no...
	}
	// HERO or VILLAIN
	console.log('I am', registeredRole);
});

.unregisterRole(callback [Function])

Unregisters a role.

NOTE 1: This is available in worker process ONLY

NOTE 2: When a worker process that has a role dies, the registered role is freed and any other worker process without a role may claim it. by calling .registerRole().

.sendToRole(roleName [String], message [Mixed])

Sends a message to a given role.

Events

cluster-mode module is also an event emitter.

cluster

Emitted when your application process starts.

This event will be called when: master process is ready, worker process is ready or non-cluster process is ready.

When master process is ready, the callback will recieve two arguments: "master.ready" and pid.

When worker process is ready, the callback will recieve two arguments: "worker.ready" and pid.

When non-cluster process is ready, the callback will recieve one argument: "non.ready".

cluster.master.ready

Emitted when your cluster master process is ready.

The callback will be passed the PID of the process.

cluster.worker.ready

Emitted when your cluster worker process is ready.

The callback will be passed the PID of the process.

cluster.non.ready

Emitted when your non-cluster process is ready.

sync

Emitted when worker map is synced.

The callback will be passed the worker map (Same map as .getWorkers() would return).

NOTE: If sync option is set to false in the configuration object for .start(), the event will NOT be emitted.

auto.spawn

Emitted when a worker is auto re-spawned.

reload

Emitted when your cluster application is reloading.

When each worker process is reloaded: the callback function will recieve three arguments: "reloading", pid, and worker id.

When reload is complete: the callback function will recieve one argument: '"complete"'.

reload.reloading

Emitted when your cluster application is reloading.

The event is emitted on each worker process reload.

Your callback function recieves two arguments: pid and worker id.

reload.complete

Emitted when your cluster process has completed realoding.

message

The event is emitted when the process recieves a message object from another cluster process by .send().

The callback will be passed the message object.

Message Object Structure:

{
	from: <worker ID>/<string "master">
	msg: <message data>
}

Example Code:

var cluster = require('cluster-mode');
cluster.on('message', function (data) {
	console.log('message was sent from', data.from);
	console.log('sent message is', data.msg);
});

exit

Emitted when your process is about to exit.

The callback of this event will recieve two arguments: code and signal.

code: 0 for expected exit and 1 for exiting with an error.

signal: If the process is exiting by a signal such as SIGINT.

NOTE: Catching this event does NOT prevent the process from exiting.