hyperlog
Merkle DAG that replicates based on scuttlebutt logs and causal linking
npm install hyperlog
var hyperlog = require('hyperlog')
var log = hyperlog(db) // where db is a levelup instance
// add a node with value 'hello' and no links
log.add(null, 'hello', function(err, node) {
console.log('inserted node', node)
// insert 'world' with a link back to the above node
log.add([node.key], 'world', function(err, node) {
console.log('inserted new node', node)
})
})
Replicate graph
To replicate this log with another one simply use log.replicate()
and pipe it together with a replication stream from another log.
var l1 = hyperlog(db1)
var l2 = hyperlog(db2)
var s1 = l1.replicate()
var s2 = l2.replicate()
s1.pipe(s2).pipe(s1)
s1.on('end', function() {
console.log('replication ended')
})
A detailed write-up on how this replication protocol works will be added to this repo in the near future. For now see the source code.
API
log = hyperlog(db, opts={})
Create a new log instance. Valid keys for opts
include:
id
- some (ideally globally unique) string identifier for the log.valueEncoding
- a levelup-style encoding string or object (e.g."json"
)hash(links, value)
- a hash function that runs synchronously. Defaults to a SHA-256 implementation.asyncHash(links, value, cb)
- an asynchronous hash function with node-style callback (cb(err, hash)
).identity
,sign
,verify
- values for creating a cryptographically signed feed. See below.
You can also pass in an identity
and sign
/verify
functions which can be
used to create a signed log:
{
identity: aPublicKeyBuffer, // will be added to all nodes you insert
sign: function (node, cb) {
// will be called with all nodes you add
var signatureBuffer = someCrypto.sign(node.key, mySecretKey)
cb(null, signatureBuffer)
},
verify: function (node, cb) {
// will be called with all nodes you receive
if (!node.signature) return cb(null, false)
cb(null, someCrypto.verify(node.key, node.signature. node.identity))
}
}
log.add(links, value, opts={}, [cb])
Add a new node to the graph. links
should be an array of node keys that this node links to.
If it doesn't link to any nodes use null
or an empty array. value
is the value that you want to store
in the node. This should be a string or a buffer. The callback is called with the inserted node:
log.add([link], value, function(err, node) {
// node looks like this
{
change: ... // the change number for this node in the local log
key: ... // the hash of the node. this is also the key of the node
value: ... // the value (as the valueEncoding type, default buffer) you inserted
log: ... // the peer log this node was appended to
seq: ... // the peer log seq number
links: ['hash-of-link-1', ...]
}
})
Optionally supply an opts.valueEncoding
.
log.append(value, opts={}, [cb])
Add a value that links all the current heads.
Optionally supply an opts.valueEncoding
.
log.batch(docs, opts={}, [cb])
Add many documents atomically to the log at once: either all the docs are inserted successfully or nothing is inserted.
docs
is an array of objects where each object looks like:
{
links: [...] // array of ancestor node keys
value: ... // the value to insert
}
The callback cb(err, nodes)
is called with an array of nodes
. Each node
is
of the form described in the log.add()
section.
You may specify an opts.valueEncoding
.
log.get(hash, opts={}, cb)
Lookup a node by its hash. Returns a node similar to .add
above.
Optionally supply an opts.valueEncoding
.
log.heads(opts={}, cb)
Get the heads of the graph as a list. A head is node that no other node links to.
log.heads(function(err, heads) {
console.log(heads) // prints an array of nodes
})
The method also returns a stream of heads which is useful if, for some reason, your graph has A LOT of heads
var headsStream = log.heads()
headsStream.on('data', function(node) {
console.log('head:', node)
})
headsStream.on('end', function() {
console.log('(no more heads)')
})
Optionally supply an opts.valueEncoding
.
changesStream = log.createReadStream([options])
Tail the changes feed from the log. Everytime you add a node to the graph the changes feed is updated with that node.
var changesStream = log.createReadStream({live:true})
changesStream.on('data', function(node) {
console.log('change:', node)
})
Options include:
{
since: changeNumber // only returns changes AFTER the number
live: false // never close the change stream
tail: false // since = lastChange
limit: number // only return up to `limit` changes
until: number // (for non-live streams) only returns changes BEFORE the number
valueEncoding: 'binary'
}
replicationStream = log.replicate([options])
Replicate the log to another one using a replication stream. Simply pipe your replication stream together with another log's replication stream.
var l1 = hyperlog(db1)
var l2 = hyperlog(db2)
var s1 = l1.createReplicationStream()
var s2 = l2.createReplicationStream()
s1.pipe(s2).pipe(s1)
s1.on('end', function() {
console.log('replication ended')
})
Options include:
{
mode: 'push' | 'pull' | 'sync', // set replication mode. defaults to sync
live: true, // keep the replication stream open. defaults to false
metadata: someBuffer, // send optional metadata as part of the handshake
frame: true // frame the data with length prefixes. defaults to true
}
If you send metadata
it will be emitted as an metadata
event on the stream.
A detailed write up on how the graph replicates will be added later.
log.on('preadd', function (node) {})
On the same tick as log.add()
is called, this event fires with the node
about to be inserted into the log. At this stage of the add process, node has
these properties:
node.log
node.key
node.value
node.links
log.on('add', function (node) {})
After a node has been successfully added to the log, this event fires with the
full node
object that the callback to .add()
gets.
log.on('reject', function (node) {})
When a node is rejected, this event fires. Otherwise the add
event will fire.
You can track preadd
events against both add
and reject
events in
combination to know when the log is completely caught up.
Hyperlog Hygiene
A hyperlog will refer to potentially many different logs as it replicates with others, each with its own ID. Bear in mind that each hyperlog's underlying leveldb contains a notion of what its own local ID is. If you make a copy of a hyperlog's leveldb and write different data to each copy, the results are unpredictable and likely disastrous. Always only use the included replication mechanism for making hyperlog copies!
License
MIT