4.4 KiB

Raw Blame History

SQLite3 Cluster

Works with node cluster, or completely and unrelated node processes.

Note: Most people would probably prefer to just use PostgreSQL rather than wrap SQLite as a service... but I am not most people.

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 use SQLite in each process.

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+).

This also works with SQLCipher.

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 sqlite = require('sqlite3-cluster');
var numCores = require('os').cpus().length;

var opts = {
  filename: '/tmp/mydb.sqlcipher'
, sock: '/tmp/mydb.sqlcipher.sock'
, verbose: false

  // 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

  // if using SQLCipher, you can supply the key and desired bit-length and the
  // appropriate PRAGMA statements will be issued before the database is returned
, key: '00000000000000000000000000000000'
, bits: 128
};

sqlite.create(opts).then(function (db) {
  // same api as new sqlite3.Database(options.filename)

  client.run("SELECT ?", ['Hello World!'], function (err) {
    if (err) {
      console.error('[ERROR]', cluster.isMaster && '0' || cluster.worker.id);
      console.error(err);
      return;
    }

    console.log('[this]', cluster.isMaster && '0' || cluster.worker.id);
    console.log(this);
  });
});

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

The API is exactly the same as node-sqlite, with these few exceptions:

1 Database Creation

Instead of this:

var db = new require('sqlite3').Database(filename);

You must do this:

require('sqlite3-cluster').create(filename);

Also, verbose becomes an option to pass into the create function, rather than a function to call. Even though verbose is passed on create, it still behaves globally across all databases.

2 db.escape

This is an additional helper function.

If you need at any time to concatonate strings with user input (which you should rarely need to do since db.run(stmt, arr, fn) is usually sufficient), you can use the escape function.

var sqlEscape = require('sqlite3-cluster').escape;

also

require('sqlite3-cluster').create(options).then(function (db) {
  // obligatory xkcd reference https://xkcd.com/327/
  var userInput = db.escape("Robert'); DROP TABLE Students;");
});

3 serialize / parallelize

db.serialize(fn) and db.parallelize(fn) are not supported because it would require copying a chunk of code from node-sqlite3 and adapting it.

It wouldn't be a difficult task, just tedious and generally no longer necessary since recent versions of node include native Promises.

Standalone / Master Mode is raw sqlite3

The master in the cluster (meaning opts.serve = true) will have a direct connection to the sqlite3 database using node-sqlite, directly.

Likewise, when only one process is being used (opts.standalone = true) the listener is not started and the connection is direct.

If you take a look at wrapper.js you'll see that it simply resolves with an instance of node-sqlite3.

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.

4.4 KiB Raw Blame History