NoSQL for the little man
Switch branches/tags
Nothing to show
Clone or download
Pull request Compare This branch is 122 commits behind felixge:master.
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.

Dirty - NoSQL for the little man

A key value store for node.js that is simple, fast & dirty.

Dirty stores all documents in memory like Redis while allowing you to iterate over them in JavaScript similar to CouchDB.

What makes dirty dirty?

  • It is the minimal viable key value store (< 250 LoC)
  • You embedded it into your actual applications
  • The db format is append only newline separated JSON
  • No network support out of the box (seriously)
  • It can only store js objects, primitives and arrays need to be wrapped in an object

The big picture

Key-value stores are stupid by design. You ask them trivial questions like "What is the value of key X?" and they are pretty fast and good with answering those.

However, beyond caching, those are not the kind of questions you want to ask your database. And you certainly don't want to find the answer by asking tons of trivial questions over a network, just to compute the final result on the client side. It's not good for your network nor a lot of fun to do.

Yes, some key-value stores will offer you some sort of "query" language or ways to pre-compute answers to frequently asked questions. But honestly, those are doomed to suffer from the same problems as relational databases or will be limited in the complexity of the questions they can answer.

How does dirty avoid those problems? By cheating. It avoids the whole network trouble by simply not shipping with a network interface. This discourages asking dirty trivial questions over a network. The need for a query language is avoided giving you access to the raw data using JavaScript.

Dirty allows people to build their very own, very fast databases that do exactly what they need and nothing else. Making those high-level databases available to your system is trivial, you just create a REST service with node. As an extra, you can also use node's non-blocking I/O to query other datasources.


Create database with 2 documents:

  Dirty = require('../lib/dirty').Dirty,
  posts = new Dirty('test.dirty');

posts.add({hello: 'dirty world!'});
posts.set('my-id', {looks: 'nice'});


$ node example1.js
$ cat test.dirty 
{"hello":"dirty world!","_id":"3b8f86684573afd0b4ceba34f7b0566e"}

Reload the database from disk

  Dirty = require('../lib/dirty').Dirty,
  posts = new Dirty('test.dirty');

  .addCallback(function() {


$ node example2.js
{"looks": "nice", "_id": "my-id"}

Filter documents from the db

  Dirty = require('../lib/dirty').Dirty,
  posts = new Dirty('test.dirty');

  .addCallback(function() {
    var docs = posts.filter(function(doc) {
      return ('hello' in doc);


$ node hello.js
[{"hello": "dirty world!", "_id": "3b8f86684573afd0b4ceba34f7b0566e"}]

That's it! No more features. Dirty is quite minimal. It encourages you to use it in a lego-style fashion.

API Documentation

new Dirty([filename, options])

Creates a new dirty database. If filename does not exist, a new file is created. If filename is not specified, dirty operates in-memory only (useful for testing).

The default options are:

  flushInterval: 10,
  flushLimit: 1000,

This means that documents are written to disk in bulks of 100 or after 10ms, whichever occurs first.


Returns how many documents are currently loaded.


Returns a promise that finishes once all records have been read from the current database.

You can already start querying or using a partially loaded database. However, documents overwrite their ids as they are loaded, so be careful.


Properly closes the database file and removes any remaining timers.

Dirty.add(doc, [callback])

Adds a new document to the database and assigns it a uuid as the id.

doc needs to be a JavaScript object. Other data types are not allowed, but can be included inside the object as far as they can be serialized to JSON.

callback is an optional function that is called once the document has been written to disk.

The function returns the uuid created for the record. It also modifies the doc object that is being passed in by adding a _id property.


var doc = {hello: 'world'};
var uuid = db.add(doc);
p(uuid); // "3b8f86684573afd0b4ceba34f7b0566e"
p(doc._id == uuid); // true

Dirty.set(id, doc, [callback])

Identical to Dirty.add() except:

  • id is used instead of a uuid
  • The function has no return value

Dirty.remove(id, [callback])

Removes given id and fires callback once the change has flushed to disk.

Important: A removed id will show up in Dirty.fiter() as {_deleted: true} until it has been removed from disk.


Returns the object for the given id or undefined if it is not set.


Calls callback for each document in the database with doc as the first parameter. Returns an array with all docs for which callback returns a trueish value.


posts.add({title: 'cool post'});
posts.add({title: 'awesome post'});

var docs = posts.filter(function(doc) {
  return doc.title.match(/awesome/);
p(docs); // [{"title": "awesome post", "_id": "2895ed039b68455e23a202627537030c"}]


Flushes all documents to disk that are only stored in memory at that point. Returns a promise that finishes once those records have been written to disk.

Dirty.addListener('flush', callback)

The 'flush' event is emitted whenever the database gets into full sync with its disk file. This is not the same event as the promise returned by Dirty.flush(), it guarantees full sync while the flush promise merely guarantees a previous state to have fully flushed.


dirty ships with a small set of benchmarks which can be invoked by running make benchmark. On my eager laptop the output usually looks like this:

$ make benchmark
node benchmark/filter.js
763000 docs added in 1030ms
flushed to disc, starting filtering ...

Filtered 14497000 docs in 1026 ms 	(14129630 per sec)
node benchmark/set.js
MEMORY: 763000 writes in 1031 ms 	(740058 per sec)
DISK:   763000 writes in 14.68 sec 	(51965 per sec)


  • filter.js adds as many records in 1000ms as possible (like set.js), and then iterates over them for 1000ms filtering out 50% of the items and outputs the results.
  • set.js uses the Dirty.set() method to add as many objects to the database as possible for 1000ms and outputs the results

That is not a very realistic load scenario and not really comparable to other stores that do much more (like networking). If anything, you can learn that I care about the performance of dirty by providing a benchmark as part of the distribution.

Make sure to read Benchmarks: You are Doing it Wrong by @janl if you're interested in doing some benchmarking of your own.


Dirty is licensed under the MIT license.