memstore-cluster
v1.0.0
Published
A wrapper to enable the use of a in-process store with node cluster via a socket server (i.e. for Raspberry Pi 2).
Downloads
14
Maintainers
Readme
memstore Cluster
A very simple in-memory object store for use with node cluster (or even completely and unrelated node processes).
Node.js runs on a single core, which isn't very effective.
You can run multiple Node.js instances to take advantage of multiple cores, but if you do that, you can't share memory between processes.
This module will either run client-server style in environments that benefit from it (such as the Raspberry Pi 2 with 4 cores), or in-process for environments that don't (such as the Raspberry Pi B and B+).
Note: Most people would probably prefer to just use Redis rather than wrap a dumb memstore as a service... but I am not most people.
Also works with level-session-store (leveldb), connect-session-knex (SQLite3), session-file-store (fs), and any other embedded / in-process store.
Usage
The default behavior is to try to connect to a master and, if that fails, to become the master.
However, if you are in fact using the cluster
rather than spinning up random instances,
you'll probably prefer to use this pattern:
var cluster = require('cluster');
var store = require('memstore-cluster');
var numCores = require('os').cpus().length;
var opts = {
sock: '/tmp/memstore.sock'
// If left 'null' or 'undefined' this defaults to a similar memstore
// with no special logic for 'cookie' or 'expires'
, store: cluster.isMaster && new require('express-session/session/memory')()
// a good default to use for instances where you might want
// to cluster or to run standalone, but with the same API
, serve: cluster.isMaster
, connect: cluster.isWorker
, standalone: (1 === numCores) // overrides serve and connect
};
memstore.create(opts).then(function (store) {
// same api as new sqlite3.Database(options.filename)
store.get(id, function (err, data) {
console.log(data);
});
// app.use(expressSession({ secret: 'keyboard cat', store: store }));
});
process.on('unhandledPromiseRejection', function (err) {
console.error('Unhandled Promise Rejection');
console.error(err);
console.error(err.stack);
throw err;
});
If you wish to always use clustering, even on a single core system, see test-cluster.js
.
Likewise, if you wish to use standalone mode in a particular worker process see test-standalone.js
.
API
This is modeled after Express' Session Store Implementation
Note: These are only exposed if the underlying store supports them.
CRUD methods
store.set(id, data, fn) => (error)
store.get(id, fn) => (error, data)
store.touch(id, data, fn) => (error)
store.destroy(id, fn) => (error)
Helpers
store.all(fn) => (error, array)
store.clear(fn) => (error)
store.length(fn) => (error, length)
See https://github.com/expressjs/session#session-store-implementation@4.x for full details
Standalone / Master Mode is in-process
The master
in the cluster (meaning opts.serve = true
) will directly hold the express-session/session/memory
memory store.
Likewise, when only one process is being used (opts.standalone = true
) the listener is
not started and API is completely in-process.
If you take a look at wrapper.js
you'll see that it's a rather simple memory store instance.
Security Warning
Note that any application on the system could connect to the socket.
In the future I may add a secret
field in the options object to be
used for authentication across processes. This would not be difficult,
it's just not necessary for my use case at the moment.