k-bucket

k-bucket

Stability: 2 - Unstable

Kademlia DHT K-bucket implementation as a binary tree.

Contributors

@tristanls, @mikedeboer, @deoxxa, @feross, @nathanph

Installation

npm install k-bucket

Tests

npm test

Usage

var KBucket = require('k-bucket');

var kBucket = new KBucket({
    localNodeId: new Buffer("my node id") // default: random data
});

Overview

A Distributed Hash Table (DHT) is a decentralized distributed system that provides a lookup table similar to a hash table.

k-bucket is an implementation of a storage mechanism for keys within a DHT. It stores contact objects which represent locations and addresses of nodes in the decentralized distributed system. contact objects are typically identified by a SHA-1 hash, however this restriction is lifted in this implementation. Additionally, node ids of different lengths can be compared.

This Kademlia DHT k-bucket implementation is meant to be as minimal as possible. It assumes that contact objects consist only of id. It is useful, and necessary, to attach other properties to a contact. For example, one may want to attach ip and port properties, which allow an application to send IP traffic to the contact. However, this information is extraneous and irrelevant to the operation of a k-bucket.

arbiter function

This k-bucket implementation implements a conflict resolution mechanism using an arbiter function. The purpose of the arbiter is to choose between two contact objects with the same id but perhaps different properties and determine which one should be stored. As the arbiter function returns the actual object to be stored, it does not need to make an either/or choice, but instead could perform some sort of operation and return the result as a new object that would then be stored. See kBucket.update(contact, index) for detailed semantics of which contact (incumbent or candidate) is selected.

For example, an arbiter function implementing a vectorClock mechanism would look something like:

// contact example
var contact = {
    id: new Buffer('contactId'),
    vectorClock: 0
};

function arbiter(incumbent, candidate) {
    if (incumbent.vectorClock > candidate.vectorClock) {
        return incumbent;
    }
    return candidate;
};

Alternatively, consider an arbiter that implements a Grow-Only-Set CRDT mechanism:

// contact example
var contact = {
    id: new Buffer('workerService'),
    workerNodes: {
        '17asdaf7effa2': { host: '127.0.0.1', port: 1337 },
        '17djsyqeryasu': { host: '127.0.0.1', port: 1338 }
    }
};

function arbiter(incumbent, candidate) {
    // we create a new object so that our selection is guaranteed to replace
    // the incumbent
    var merged = {
        id: incumbent.id, // incumbent.id === candidate.id within an arbiter
        workerNodes: incumbent.workerNodes
    };

    Object.keys(candidate.workerNodes).forEach(function (workerNodeId) {
        merged.workerNodes[workerNodeId] = candidate.workerNodes[workerNodeId];
    });

    return merged;
}

Notice that in the above case, the Grow-Only-Set assumes that each worker node has a globally unique id.

Documentation

KBucket

Implementation of a Kademlia DHT k-bucket used for storing contact (peer node) information.

For a step by step example of k-bucket operation you may find the following slideshow useful: Distribute All The Things.

KBucket starts off as a single k-bucket with capacity of k. As contacts are added, once the k+1 contact is added, the k-bucket is split into two k-buckets. The split happens according to the first bit of the contact node id. The k-bucket that would contain the local node id is the "near" k-bucket, and the other one is the "far" k-bucket. The "far" k-bucket is marked as don't split in order to prevent further splitting. The contact nodes that existed are then redistributed along the two new k-buckets and the old k-bucket becomes an inner node within a tree data structure.

As even more contacts are added to the "near" k-bucket, the "near" k-bucket will split again as it becomes full. However, this time it is split along the second bit of the contact node id. Again, the two newly created k-buckets are marked "near" and "far" and the "far" k-bucket is marked as don't split. Again, the contact nodes that existed in the old bucket are redistributed. This continues as long as nodes are being added to the "near" k-bucket, until the number of splits reaches the length of the local node id.

As more contacts are added to the "far" k-bucket and it reaches its capacity, it does not split. Instead, the k-bucket emits a "ping" event (register a listener: kBucket.on('ping', function (oldContacts, newContact) {...}); and includes an array of old contact nodes that it hasn't heard from in a while and requires you to confirm that those contact nodes still respond (literally respond to a PING RPC). If an old contact node still responds, it should be re-added (kBucket.add(oldContact)) back to the k-bucket. This puts the old contact on the "recently heard from" end of the list of nodes in the k-bucket. If the old contact does not respond, it should be removed (kBucket.remove(oldContact)) and the new contact being added now has room to be stored (kBucket.add(newContact)).

Public API

• KBucket.distance(firstId, secondId)
• new KBucket(options)
• kBucket.add(contact, [bitIndex])
• kBucket.closest(contact, n, [bitIndex])
• kBucket.count()
• kBucket.get(id, [bitIndex])
• kBucket.remove(contact, [bitIndex])
• kBucket.toArray()
• Event 'ping'

KBucket.distance(firstId, secondId)

• firstId: Buffer Buffer containing first id.
• secondId: Buffer Buffer containing second id.
• Return: Integer The XOR distance between firstId and secondId.

Finds the XOR distance between firstId and secondId.

new KBucket(options)

• options:
  • arbiter: Function (Default: vectorClock arbiter) function (incumbent, candidate) { return contact; } An optional arbiter function that givent two contact objects with the same id returns the desired object to be used for updating the k-bucket. For more details, see arbiter function.
  • localNodeId: Buffer An optional Buffer representing the local node id. If not provided, a local node id will be created via crypto.randomBytes(20).
  • numberOfNodesPerKBucket: Integer (Default: 20) The number of nodes that a k-bucket can contain before being full or split.
  • numberOfNodesToPing: Integer (Default: 3) The number of nodes to ping when a bucket that should not be split becomes full. KBucket will emit a ping event that contains numberOfNodesToPing nodes that have not been contacted the longest.
  • root: Object CAUTION: reserved for internal use Provides a reference to the root of the tree data structure as the k-bucket splits when new contacts are added.

Creates a new KBucket.

kBucket.add(contact, [bitIndex])

• contact: Object The contact object to add.
  • id: Buffer Contact node id.
  • Any satellite data that is part of the contact object will not be altered, only id is used.
• bitIndex: Integer (Default: 0) CAUTION: reserved for internal use The bit index to which bit to check in the id Buffer.
• Return: Object The k-bucket itself.

Adds a contact to the k-bucket.

kBucket.closest(contact, n, [bitIndex])

• contact: Object The contact object to find closest contacts to.
  • id: Buffer Contact node id.
  • Any satellite data that is part of the contact object will not be altered, only id is used.
• n: Integer The maximum number of closest contacts to return.
• bitIndex: Integer (Default: 0) CAUTION: reserved for internal use The bit index to which bit to check in the id Buffer.
• Return: Array Maximum of n closest contacts to the contact.

Get the n closest contacts to the provided contact. "Closest" here means: closest according to the XOR metric of the contact node id.

kBucket.count()

• Return: Number The number of contacts held in the tree

Counts the total number of contacts in the tree.

kBucket.determineBucket(id, [bitIndex])

CAUTION: reserved for internal use

• id: Buffer Id to compare localNodeId with.
• bitIndex: Integer (Default: 0) The bit index to which bit to check in the id Buffer.
• Return: Integer -1 if id at bitIndex is 0, 1 otherwise.

Determines whether the id at the bitIndex is 0 or 1. If 0, returns -1, else 1.

kBucket.get(id, [bitIndex])

• id: Buffer The ID of the contact to fetch
• bitIndex: Integer (Default: 0) CAUTION: reserved for internal use The bit index to which bit to check in the id Buffer.
• Return: Object The contact if available, otherwise null

Retrieves the contact.

kBucket.indexOf(contact)

CAUTION: reserved for internal use

• contact: Object The contact object.
  • id: Buffer Contact node id.
  • Any satellite data that is part of the contact object will not be altered, only id is used.
• Return: Integer Index of contact if it exists, -1 otherwise.

Returns the index of the contact if it exists, returns -1 otherwise.

_NOTE: kBucket.indexOf(contact) does not use arbiter in the comparison.

kBucket.remove(contact, [bitIndex])

• contact: Object The contact object to remove.
  • id: Buffer contact node id.
  • Any satellite data can be part of the contact object, only id is used
• bitIndex: Integer (Default: 0) CAUTION: reserved for internal use The bit index to which bit to check in the id Buffer.
• Return: Object The k-bucket itself.

Removes the contact.

kBucket.splitAndAdd(contact, [bitIndex])

CAUTION: reserved for internal use

• contact: Object The contact object to add.
  • id: Buffer Contact node id.
  • Any satellite data that is part of the contact object will not be altered, only id is used.
• bitIndex: Integer (Default: 0) The bit index to which bit to check in the id Buffer.
• Return: Object The k-bucket itself.

Splits the bucket, redistributes contacts to the new buckets, and marks the bucket that was split as an inner node of the binary tree of buckets by setting self.bucket = undefined. Also, marks the "far away" bucket as dontSplit.

kBucket.toArray()

• Return: Array All of the contacts in the tree, as an array

Traverses the tree, putting all the contacts into one array.

kBucket.update(contact, index)

CAUTION: reserved for internal use

• contact: Object The contact object to update.
  • id: Buffer Contact node id
  • Any satellite data that is part of the contact object will not be altered, only id is used.
• index: Integer The index in the bucket where contact exists (index has already been computed in previous calculation).

Updates the contact by using the arbiter function to compare the incumbent and the candidate. If arbiter function selects the old contact but the candidate is some new contact, then the new contact is abandoned. If arbiter function selects the old contact and the candidate is that same old contact, the contact is marked as most recently contacted (by being moved to the right/end of the bucket array). If arbiter function selects the new contact, the old contact is removed and the new contact is marked as most recently contacted.

Event: 'ping'

• oldContacts: Array The array of contacts to ping.
• newContact: Object The new contact to be added if one of old contacts does not respond.

Emitted every time a contact is added that would exceed the capacity of a don't split k-bucket it belongs to.

Sources

The implementation has been sourced from:

• A formal specification of the Kademlia distributed hash table
• Distributed Hash Tables (part 2)