- No worker code modification is necessary.
- Provides common solution for master process.
- Maintains specified quantity of running workers.
- Runs groups of workers on the different ports for 3rd party load balancing (nginx or haproxy, for example).
- Allows configuration via JSON, JS or anything that can be
require
d out of the box. - Zero downtime successive workers' restart.
- Simple and flexible API for building extensions and development of custom master-workers solutions.
In [email protected]
we dropped support for node<4
.
If you desperately need to make it run on older node
versions, use [email protected]
.
Install luster
module and save it as runtime dependency:
$ npm install --save luster
Write minimal required configuration file for luster:
$ echo '{ "app" : "worker.js" }' > ./luster.conf.json
Run the cluster:
$ ./node_modules/.bin/luster
Read configuration manual to know more about luster features.
Following example written in plain JavaScript, not JSON, so you can name it
luster.conf.js
to launch luster without options,
or pass the configuration file path as the first argument to the script:
$ ./node_modules/.bin/luster ./configs/my_luster_configuration.js
Internally, luster tries to call the require()
in the following way:
require(path.resolve(process.cwd(), process.argv[2] || './luster.conf'));
module.exports = {
// required, absolute or relative path to configuration file
// of worker source file
app : "./worker.js",
// workers number
// number of cpu threads is used by default
workers : 4,
// options to control workers startup and shutdown processes
control : {
// time to wait for 'online' event from worker
// after spawning it (in milliseconds)
forkTimeout : 3000,
// time to wait for 'exit' event from worker
// after disconnecting it (in milliseconds)
stopTimeout : 10000,
// if worker dies in `exitThreshold` time (in milliseconds) after start,
// then its' `sequentialDeaths` counter will be increased
exitThreshold : 5000,
// max allowed value of `sequentialDeaths` counter
// for each worker; on exceeding this limit worker will
// be marked as `dead` and no more automatic restarts will follow.
allowedSequentialDeaths : 10,
// if falsy, worker is considered ready after 'online' event
// it happens between forking worker and executing it
// if truly, worker is considered ready
// when you call require('luster').ready inside of it
// notice that it's only affect startup/restart logic
// worker will start handling requests right after you call 'listen' inside of it
triggerReadyStateManually : false
},
// use "server" group if you want to use web workers
server : {
// initial port for the workers;
// can be tcp port number or path to the unix socket;
// if you use unix sockets with groups of the workers,
// then path must contain '*' char, which will be replaced
// with group number
//
// worker can get port number to listen from the environment variable
// `port`, for example:
// > server.listen(process.env.port)
port : 8080,
// number of workers' groups; each group will
// have its own port number (port + group number)
groups : 2
},
// extensions to load
// each key in the "extensions" hash is a npm module name
extensions : {
// luster-log-file extension example
"luster-log-file" : {
stdout : "/var/log/luster/app.stdout.log",
stderr : "/var/log/luster/app.stderr.log"
},
// luster-guard extension example
"luster-guard" : {
include: [ '**/*.js' ],
exclude: [ '**/node_modules/**' ]
}
},
// if extensions' modules can't be resolved as related to
// luster module or worker path, then absolute path
// to the directory, which contains extensions modules
// must be declared here:
extensionsPath : "/usr/local/luster-extensions",
// max time to wait for extensions initialization
extensionsLoadTimeout : 10000,
// if your app or used extensions extensively use luster
// internal events then you can tweak internal event emitters
// listeners number limit using following option.
// default value is `100`, option must be a number else EventEmitter
// throws an error on configuration.
maxEventListeners : 100
};
Extensions is a simple Node.js module, which must export object with configure
function,
which will be called during master and worker configuration.
Synchronous extension initialization:
module.exports = {
configure : function(config, clusterProcess) {
// has `get` method:
// var someProp = config.get('some.property.path', defaultValue);
this.config = config;
if (clusterProcess.isMaster) {
this.initializeOnMaster(clusterProcess);
} else {
this.initializeOnWorker(clusterProcess);
}
}
}
Asynchronous extension initalization:
module.exports = {
initializeOnMaster : function(master, done) {
// emulate async operation
setTimeout(function() {
// do something
done();
}, 500);
},
initializeOnWorker : function(worker, done) {
// emulate async operation
setTimeout(function() {
// do something
done();
}, 300);
},
configure : function(config, clusterProcess, done) {
// has `get` method:
// var someProp = config.get('some.property.path', defaultValue);
this.config = config;
if (clusterProcess.isMaster) {
this.initializeOnMaster(clusterProcess, done);
} else {
this.initializeOnWorker(clusterProcess, done);
}
}
}
To enable asynchronous initalization of an extension, configure
function must be declared with 3 or more arguments,
where 3-rd argument is callback, which must be called by extensions when initialization has been finished.
Callback accepts one optional argument: an error, if initalization failed.
If you are somehow lost in how master-worker interaction works, feel free to use NODE_DEBUG=luster:eex
when launching your app.
For example, you can check it within luster examples
folder:
cd examples/custom_master_and_ipc/
NODE_DEBUG=luster:eex npm run start
You will see the sequence of events both on master and workers, along with underlying IPC messages.