Fork me on GitHub


An Evented I/O B-tree for Node.js.

Strata is part of a collection of database primitives that you can use to design your own distributed databases for your Node.js applications. I call this collection the Strata Universe. It culimates in Locket, a pure-JavaScript implementation of LevelDB.

Strata is a concurrent, b‑tree primitive, in pure-JavaScript for Node.js.

A b‑tree is a data structure used by databases to store records organized in large pages on disk.

By concurrent I mean that multiple queries can make progress on a descent of the b‑tree. Multiple reads can all navigate the b‑tree simultaneously. Multiple reads can also make progress in the presence of a write, so long as they are not reading a page that is being written. This is the equivalence to "threading" in other database engines, but evented for Node.js.

Strata is a database primitive, it is not supposed to be used a as a general purpose database by it's lonesome. Strata an implementation of a b‑tree and it's interface exposes b‑concepts. If you're using Strata you're either implementing a database engine, or your taking your indexes and queries into your own hands.


Install from NPM.

npm install b-tree

B-Tree Properties

TK: Unique keys, but duplicate keys are super easy to fake.

Creating a B-Tree

You must create the b‑tree object first specifying a directory where records are stored.

function createOpenClose (directory, callback) {
    var strata


    function create () {
        strata = new Strata(directory)

    function created (error) {
        if (error) return callback(error)


    function reopen (error) {
        if (error) return callback(error)

        strata = new Strata(directory)

    function open (error) {
        if (error) return callback(error)


createOpenClose(function (error) { if (error) throw error })

Extractor and Comparator

Strata will store any JSON object or value. By default it compares using the JavaScript comparison operators < and >. If you're string a String or a Number the default comparisions will work just fine.

If you're storing objects you're going to want to order by one or more of the object's properties. To do this you specify an extractor property to extract the key properties from the object and a comparator to compare the extracted key.

function extractor (person) {
    return {
        firstName: person.firstName,
        lastName: person.lastName

function comparator (left, right) {
    var compare = left.lastName < right.lastName
                ? -1 : left.lastName > right.lastName
                     ? 1 : 0
    if (compare != 0) {
        return compare
    return left.firstName < right.firstName
         ? -1 : left.firstName > right.firstName
              ? 1 : 0

var strata = new Strata({ extractor: extractor, comparator: comparator })

Note that neither the extractor nor the comparator are stored in the Strata tree. You must provide them when you construct a Strata tree object both when you are creating a new Strata tree or opening an exiting one. That is, you need to know the type of data stored in the tree.

Because Strata is not a general purpose data store, it is a b‑tree privitive.

Searching the B‑Tree

With Strata you either create read-only iterator, or a read/write mutator. The mutator is a superset of the iterator so let's start there.

// Returns true if the key exists in the tree.
function hasKey (strata, sought, callback) {
    var found

    strata.iterator(sought, cursor)

    function cursor (error, cursor) {
        if (error) callback(error)
        else {
            found = cursor.index >= 0

    function unlocked (error) {
        if (error) callback(error)
        else callback(null, found)

hasKey(strata, 'c', function (error, exists) {
    if (error) throw error
    if (exists) console.log('I found it.')

In the above, we create a read-only Cursor using the Strata.iterator function. That returns an iterator that holds a shared lock on the leaf page that either contains the records for the given key, or else would contain the record for the given key if it existed in the leaf page. The Cursor says that the record is here if it is in the tree, or it should go here if it's not.

The default Strata b‑tree will have defaults for branch size and leaf size, but you can adjust those as well.

Properties to the constructor...

new Strata(location[, options]).

Constructs a new b-tree that stores its files in the directory provided by location. It does not open or close the b‑tree.


new Strata() takes an optional options object as its second argument; the following properties are accepted:

Opens the b-tree.


Creates a new, empty b-tree. It will raise an exception if there is anything in the location directory.

Searching and Editing

You search and edit the b‑ separate from editing it.

If the Cursor.index property is zero or more, it is the index of the record in the leaf page. If the Cursor.index property is less than zero, then it's compliment is the index of where the record should go in the leaf page.

In the hasKey function above we simply return whether or not the record exists based on the cursor index.

Scanning the B-Tree

Ed: The following is stupid wrong and stupid.

var range = cadence(function (async, strata, start, stop) {
  strata.iterator(start, check(atLeaf));

  function atLeaf (cursor) {
    fetch(cursor.index &lt; 0 ? ~cursor.index : cursor.index);

    function fetch (index) {
      if (index &lt; cursor.length) {
        cursor.get(index, check(push));
      } else {;

    function push (record) {
      if (record &lt; stop) {
        fetch(index + 1);
      } else {

    function advanced (success) {
      if (success) done();
      else fetch(0);

    function done () {
      callback(null, found);

range(strata, 'c', 'i', function (error, found) {
  if (error) throw error;

A Note on Examples

I use a control-flow library of my own creation called Cadence that I'm just crazy about.

The primary benefit of Cadence it asynchronous try/catch error handling. This has made it very easy to create deeply nested asynchronous operations, yet have errors propagate up to the caller, the context that comes from maintaining a stack and unwinding it on error.

Secondary benefit is that Cadence always uses a trampoline. This allows me to cache aggressively without having to worry about blowing the stack.

Finally, Cadence is pure-JavaScript and old fashioned JavaScript. It doesn't depend on language features that have yet to drop. No transpilers.