Generate 128-bit unique identifiers for various specifications. In particular:
- ULID
- Monotonic ULID
- UUID 1 (Variant 1 Version 1)
- UUID 4 (Variant 1 Version 4)
- UUID 6 (Variant 1 Version 6)
- Nil UUID (Variant 0 Version 0)
- Uuid (Unknown Variant and Version)
const {
Ulid,
UlidMonotonic,
Uuid,
Uuid1,
Uuid4,
Uuid6,
UuidNil,
idCompare,
idEqual,
} = require('id128');
// Id factories
[
Ulid,
UlidMonotonic,
Uuid1,
Uuid4,
Uuid6,
UuidNil,
].forEach((IdType) => {
// Identify the factory
console.log(IdType.name);
// Generate a new id
const id = IdType.generate();
// Get the smallest valid id
const min = IdType.MIN();
// Get the largest valid id
const max = IdType.MAX();
// Type-check the id
console.log(id instanceof IdType.type)
// Compare ids
console.log(id.equal(id));
console.log(! id.equal(min));
console.log(! id.equal(max));
console.log(id.compare(min) === 1);
console.log(id.compare(id) === 0);
console.log(id.compare(max) === -1);
// Encode the id in its canonical form
const canonical = id.toCanonical();
console.log(canonical);
// Encode the id for efficient db storage
const raw = id.toRaw();
console.log(raw);
// Verify a canonically formatted id
console.log(IdType.isCanonical(canonical));
// Decode a valid canonically formatted id
console.log(id.equal(IdType.fromCanonical(canonical)));
// Decode a canonically formatted id, skipping validation
console.log(id.equal(IdType.fromCanonicalTrusted(canonical)));
// Verify a raw formatted id
console.log(IdType.isRaw(raw));
// Decode a valid raw formatted id
console.log(id.equal(IdType.fromRaw(raw)));
// Decode a raw formatted id, skipping validation
console.log(id.equal(IdType.fromRawTrusted(raw)));
});
// Uuid Factory
[0, 1, 4, 6].forEach((version) => {
// Generate a new id
const id = Uuid.generate({ version });
// Get the smallest valid id
const min = Uuid.MIN({ version });
// Get the largest valid id
const max = Uuid.MAX({ version });
// Type-check the id
console.log(id instanceof Uuid.type)
// Encode the id in its canonical form
const canonical = id.toCanonical();
console.log(canonical);
// Encode the id for efficient db storage
const raw = id.toRaw();
console.log(raw);
// Decode a valid canonically formatted id
console.log(id.equal(Uuid.fromCanonical(canonical)));
// Decode a canonically formatted id, skipping validation
console.log(id.equal(Uuid.fromCanonicalTrusted(canonical)));
// Decode a valid raw formatted id
console.log(id.equal(Uuid.fromRaw(raw)));
// Decode a raw formatted id, skipping validation
console.log(id.equal(Uuid.fromRawTrusted(raw)));
});
// Static Utilities
// Equate arbitrary ids
console.log(idEqual(Ulid.generate(), Uuid4.generate()))
// Compare arbitrary ids
console.log(idCompare(Ulid.generate(), Uuid4.generate()))
Return the name of the generated id type.
Return the type of the generated id instances for type-checking
with the instanceof
operator.
Return a new id instance without validating the bytes.
Return a new id instance.
Return the id instance with the smallest valid value.
Return the id instance with the largest valid value.
Decode an id from its canonical representation.
Throw InvalidEncoding
if the string is undecodable.
Decode an id from its canonical representation. Skip validation and assume the input is decodable.
Decode an id from its raw representation.
Throw InvalidEncoding
if the string is undecodable.
Decode an id from its raw representation. Skip validation and assume the input is decodable.
Encode the given id in the canonical form.
Throw InvalidBytes
if the id is not 128-bit conformant.
Encode the given id in the raw form.
Throw InvalidBytes
if the id is not 128-bit conformant.
Verify if a string is a valid canonical encoding.
Verify if a string is a valid raw encoding.
Return the actual byte array representing the id.
Return a new instance of the id with the same bit signature.
Determine how this id is ordered against another.
Determine if this id has the same bytes as another.
Encode this id in its canonical form.
Encode this id in its raw form.
Determine if the left id is less than | equal to | greater than
the right id using lexicographical byte order.
Determine if 2 ids have the same byte value.
const { Ulid } = require('id128');
Ulid, as specified, has some nice properties:
- collision resistant: 80-bits of randomness
- k-ordered: prefixed with millisecond precision timestamp
- database friendly: fits within a uuid and generally appends to the index
- human friendly: canonically encodes as a case-insensitive Crockford 32 number
It is useful when you need a distributed domain unique id.
Return a Date object for the epoch milliseconds encoded in the id.
Return a new id instance. Set any argument to null
or undefined
to trigger
its default behavior.
time
defaults to the current time. It can be given either as a Date
object
or epoch milliseconds (milliseconds since January 1st, 1970).
Throw InvalidEpoch
for times before the epoch or after approximately August 2nd, 10889.
This is provided mostly for unit tests.
Format tttt tttt tttt rrrr rrrr rrrr rrrr rrrr
where:
t
is 4 bits of timer
is 4 bits of random
const { UlidMonotonic } = require('id128');
UlidMonotonic is inspired by the specification:
- collision resistant: 15-bits of random seeded clock sequence plus 64-bits of randomness
- total ordered: prefixed with millisecond precision timestamp plus 15-bit clock sequence
- database friendly: fits within a uuid and generally appends to the index
- human friendly: canonically encodes as a case-insensitive Crockford 32 number
It is useful when you need to guarantee a process unique id.
Return a Date object for the epoch milliseconds encoded in the id.
Return a new id instance. Set any argument to null
or undefined
to trigger
its default behavior.
time
defaults to the current time. It can be given either as a Date
object
or epoch milliseconds (milliseconds since January 1st, 1970). Extra caution is
required since setting a future time and subsequently calling generate
guarantees usage of the clock sequence.
Throw InvalidEpoch
for times before the epoch or after approximately August 2nd, 10889.
Throw ClockSequenceOverflow
when the clock sequence is exhausted.
This is provided mostly for unit tests.
Return the clock sequence to its starting position. This is provided mostly for unit tests.
Format tttt tttt tttt cccc rrrr rrrr rrrr rrrr
where:
t
is 4 bits of timec
is 4 bits of random-seeded clock sequencer
is 4 bits of random
More specifically, the clock sequence is a counter. When the first id for a new timestamp is generated, the clock sequence is seeded with random bits and the left-most clock sequence bit is set to 0, reserving 2^15 clock ticks. Whenever a time from the past seeds the generator, the previous id's time and clock sequence are used instead, with the clock sequence incremented by 1. This guarantees strict local monotonicity and preserves lexical ordering and general randomness.
Without a seeded time, UlidMonotonic is unlikely to exceed the clock sequence (the clock sequence supports generating a new id every 31 nanoseconds). However, in the unlikely event of an overflow, id generation is aborted.
const { Uuid1 } = require('id128');
Uuid1 implements the RFC 4122 time specification:
- time-based: encodes the current millisecond timestamp
- location-based: encodes the mac address of the machine
While this mostly adheres to the spec, there are a few nuances in the handling of time. Instead of encoding time as 100-nanoseconds since the Gregorian epoch, 48 bits encode milliseconds since the Gregorian epoch time and 12 bits count past time collisions, resetting whenever given a new future time. There are a few benefits:
- high precision time is unreliable in the browser so this ensures better precision
- the max supported date is now around the year 10502 instead of around 5236
- generating 4096 ids/ms (~4,000,000 ids/s) is wildly unlikely in real world uses
- in the rare hi-res overflow, the count simply spills over to the clock sequence
Return the clock sequence encoded in the id.
Return the number of prior ids generated while time stood still.
Return the MAC address encoded in the id.
Return a Date object for the epoch milliseconds encoded in the id.
Return the variant as encoded in the id. Should be 1.
Return the version as encoded in the id. Should be 1.
Return a new id instance. Set any argument to null
or undefined
to trigger
its default behavior.
time
defaults to the current time. It can be given either as a Date
object
or Gregorian milliseconds (milliseconds since October 15th, 1582). Extra caution
is required since setting a future time and subsequently calling generate
guarantees usage of the hi-res counter and clock sequence.
Throw InvalidEpoch
for times before the Gregorian epoch or after approximately May 17, 10502.
This is provided mostly for unit tests.
node
defaults to the MAC address, or a random multicast address when the MAC
address is unavailable. It can be given as an array of 6 bytes.
Return the hi-res counter to its starting position and generate a new random clock sequence seed. This is provided mostly for unit tests.
Format llll lnnn mmmm vhhh tccc aaaa aaaa aaaa
where:
l
is 4 bits of low millisecond timen
is 4 bits of hi-res timem
is 4 bits of mid millisecond timev
is 4 bits of the versionh
is 4 bits of high millisecond timet
is 2 bits of the variant followed by 2 bits of the clock sequencec
is 4 bits of the clock sequencea
is 4 bits of the machine address
const { Uuid4 } = require('id128');
Uuid4 implements the RFC 4122 random uuid specification:
- 122 random bits
- 2 bits reserved for the variant (1)
- 4 bits reserved for the version (4)
It is useful when you need a well-supported globally unique id.
Return the variant as encoded in the id. Should be 1.
Return the version as encoded in the id. Should be 4.
Format rrrr rrrr rrrr vrrr trrr rrrr rrrr rrrr
where:
r
is 4 bits of randomv
is 4 bits of the versiont
is 2 bits of the variant followed by 2 bits of random
const { Uuid6 } = require('id128');
Uuid6 implements this controversial blog post:
- time-based: encodes the current millisecond timestamp
- location-based: encodes the mac address of the machine
This is essentially the same implementation as Uuid1, however the time bits are arranged in lexicographical order. If you're looking for a spacial UUID that is optimized for clustered indices, consider Uuid6 as a viable option.
Return the clock sequence encoded in the id.
Return the number of prior ids generated while time stood still.
Return the MAC address encoded in the id.
Return a Date object for the epoch milliseconds encoded in the id.
Return the variant as encoded in the id. Should be 1.
Return the version as encoded in the id. Should be 6.
Return a new id instance. Set any argument to null
or undefined
to trigger
its default behavior.
time
defaults to the current time. It can be given either as a Date
object
or Gregorian milliseconds (milliseconds since October 15th, 1582). Extra caution
is required since setting a future time and subsequently calling generate
guarantees usage of the hi-res counter and clock sequence.
Throw InvalidEpoch
for times before the Gregorian epoch or after approximately May 17, 10502.
This is provided mostly for unit tests.
node
defaults to the MAC address, or a random multicast address when the MAC
address is unavailable. It can be given as an array of 6 bytes.
Return the hi-res counter to its starting position and generate a new random clock sequence seed. This is provided mostly for unit tests.
Format mmmm mmmm mmmm vnnn tccc aaaa aaaa aaaa
where:
m
is 4 bits of millisecond timev
is 4 bits of the versionn
is 4 bits of hi-res timet
is 2 bits of the variant followed by 2 bits of the clock sequencec
is 4 bits of the clock sequencea
is 4 bits of the machine address
const { UuidNil } = require('id128');
UuidNil implements the RFC 4122 nil uuid specification:
- 128 bits of glorious 0
It is useful as placeholder for other 128-bit ids.
Return the variant as encoded in the id. Should be 0.
Return the version as encoded in the id. Should be 0.
Format 0000 0000 0000 v000 t000 0000 0000 0000
where:
0
is 4 bits of 0v
is 4 bits of the version (also 0)t
is 2 bits of the variant (also 0) followed by 2 bits of 0
const { Uuid } = require('id128');
Uuid is a factory for generating and decoding UUIDs when the version is unknown until runtime. If the version is supported, it will produce UUIDs of the appropriate type. In exchange for the runtime flexibility, there is a necessary performance degradation. It is recommended to use this for decoding data from uncontrolled sources rather than generating new ids.
Uuid supports all the same methods as the other ID factories. All modifications to typical behavior are noted below.
Return the factories of all the supported ids.
Return a new versioned id instance without validating the bytes. Return a Uuid if an appropriate version does not exist.
Return a new versioned id instance. All additional arguments are passed through
to the associated version.
Throw UnsupportedVersion
if no associated version exists.
Return the versioned id instance with the smallest valid value.
Throw UnsupportedVersion
if no associated version exists.
Return the versioned id instance with the largest valid value.
Throw UnsupportedVersion
if no associated version exists.
Decode a versioned id from its canonical representation.
Return a Uuid if an appropriate version does not exist.
Throw InvalidEncoding
if the string is undecodable.
Decode a versioned id from its canonical representation. Return a Uuid if an appropriate version does not exist. Skip validation and assume the input is decodable.
Decode a versioned id from its raw representation.
Return a Uuid if an appropriate version does not exist.
Throw InvalidEncoding
if the string is undecodable.
Decode a versioned id from its raw representation. Return a Uuid if an appropriate version does not exist. Skip validation and assume the input is decodable.
const { Exception } = require('id128');
All exceptions are namespaced under Exception
for clarity.
const { Exception: { Id128Error } } = require('id128');
Base exception class for generic error catching.
const { Exception: { ClockSequenceOverflow } } = require('id128');
Incrementing the clock sequence is impossible. Should not happen unless manually seeding #generate
.
const { Exception: { InvalidBytes } } = require('id128');
Encoding something other than 16 bytes. Likely to happen when encoding untrusted user input.
const { Exception: { InvalidEncoding } } = require('id128');
Decoding an invalid format or non-string object. Likely to happen when decoding untrusted user input.
const { Exception: { InvalidEpoch } } = require('id128');
Generating an id with an invalid timestamp. Should not happen unless manually seeding #generate
.
const { Exception: { UnsupportedVersion } } = require('id128');
Failed to find a factory for the desired version. Likely to happen when decoding untrusted user input.
This module supports browser compilation though Webpack/Browserify with a few caveats:
- Random number generation is optimized for memory usage over speed since only a handful of ids are likely to be generated during a user's session so the overhead of generating a page of random values has poor amortized cost.
- The browser must have native support for
crypto
.Math.random
is far too insecure to support as a fallback, especially since the fallback only makes sense for older browsers with proven security holes.msCrypto
is not a supported fallback due to many of the other required features. - The browser must support:
- classes
- closures
const
andlet
Uint8Array
Symbol
This library is intended for modern browsers that keep pace with Javascript's growing ecosystem. I philosophically object to supporting efforts of companies to pour more money into broken browsers that only cause headaches for developers to support. I expect these caveats to be unnecessary within the next 5 years.
All that said, please notify me of any issues with modern browsers and I'll do my best to support you.
This module includes Typescript bindings for all primary usage patterns. I'd like to highlight some design decisions:
Each factory is exported as an instance using the same name as the type of id it produces. In Javascript, this is desirable as it provides a uniform interface regardless of the implementation. However, this complicates the Typescript type imports.
For simple cases, like constructing an id and passing it around the program, this will behave exactly as desired:
import { Ulid } from 'id128'
const id: Ulid = Ulid.generate()
When you need to check the type of the id, you should use the type
attribute:
import { Ulid } from 'id128'
const id: Ulid = Ulid.generate()
if (id instanceof Ulid.type) { ... }
If you wish to pass around the factory itself, you can import the factory type:
import { Ulid } from 'id128'
import type { UlidFactory } from 'id128'
function doSomething(factory: UlidFactory) { ... }
doSomething(Ulid)
Finally, if you need to operate on any id or id factory, you can import base types:
import type { Id, AnyIdFactory } from 'id128'
function makeOne(factory: AnyIdFactory): Id {
return factory.generate()
}
Exception classes are designed to be checked using instanceof
. Unfortunately,
Typescript broke instanceof
Error
support for a more compliant compilation.
Fortunately, the included exceptions bypass the issues caused by inheriting from
the native Error
by never overriding the constructor and implementing name
as a readonly getter, As a consequence, the exceptions actually violate the
standard Error
interface, but they fulfill the standard Function
interface.
Therefore, you can safely use instanceof
as intended:
import { UlidMonotonic } from 'id128'
import { Exception } from 'id128'
try { UlidMonotonic.generate() }
catch (err) {
if (err instanceof Exception.ClockSequenceOverflow ) { ... }
}
Originally, I was looking for an id that is independent of the database, but plays nice with database indices and data types. Most databases have built-in support for storing UUIDs efficiently, but UUID v4 does not cluster well and the other UUIDs require bit manipulation to get good performance, which will likely cause future maintenance headaches.
After a bit of research, ULID was determined to nicely solve the problem. However, the javascript implementation had 2 major issues:
- lacks database support
- it's slow, which in a single-threaded Node server is deadly
I considered sending in a patch, however I saw an opportunity for a more expressive interface, which is typically a bit harder to modify once a project is in wide use. There was also a clear pattern for encoding 128-bit ids into various formats, which seems generally useful.
Ultimately, this library strives to be:
- secure: uses cryptographic randomness to ensure general uniqueness
- performant: currently one of the fastest id generators available
- maintainable: heavily tested isolated code with a consistent interface
- extensible: modular design to easily add new ids and new encodings
To run the tests:
npm install
npm run test-all
Competitive benchmarks have been moved to benchmark-guid
To run the benchmarks:
npm install
npm run benchmark
Platform info:
==============
Darwin 18.2.0 x64
Node.JS: 15.0.0
V8: 8.6.395.16-node.15
Intel(R) Core(TM) i7-4578U CPU @ 3.00GHz × 4
Ulid
====
generate: (4,590,833rps) (avg: 217ns)
MIN: (12,491,186rps) (avg: 80ns)
MAX: (12,669,223rps) (avg: 78ns)
fromCanonical: (1,707,717rps) (avg: 585ns)
fromCanonicalTrusted: (2,078,278rps) (avg: 481ns)
fromRaw: (1,483,373rps) (avg: 674ns)
fromRawTrusted: (1,979,964rps) (avg: 505ns)
toCanonical: (3,256,155rps) (avg: 307ns)
toRaw: (6,012,244rps) (avg: 166ns)
UlidMonotonic
=============
generate: (3,787,685rps) (avg: 264ns)
MIN: (6,306,928rps) (avg: 158ns)
MAX: (6,301,217rps) (avg: 158ns)
fromCanonical: (1,423,104rps) (avg: 702ns)
fromCanonicalTrusted: (1,722,958rps) (avg: 580ns)
fromRaw: (1,381,296rps) (avg: 723ns)
fromRawTrusted: (1,698,639rps) (avg: 588ns)
toCanonical: (3,205,394rps) (avg: 311ns)
toRaw: (5,774,288rps) (avg: 173ns)
Uuid1
=====
generate: (4,984,699rps) (avg: 200ns)
MIN: (12,888,384rps) (avg: 77ns)
MAX: (12,817,435rps) (avg: 78ns)
fromCanonical: (1,226,007rps) (avg: 815ns)
fromCanonicalTrusted: (1,578,429rps) (avg: 633ns)
fromRaw: (1,306,295rps) (avg: 765ns)
fromRawTrusted: (1,626,095rps) (avg: 614ns)
toCanonical: (5,859,714rps) (avg: 170ns)
toRaw: (5,973,139rps) (avg: 167ns)
Uuid4
=====
generate: (6,492,849rps) (avg: 154ns)
MIN: (6,400,528rps) (avg: 156ns)
MAX: (6,617,714rps) (avg: 151ns)
fromCanonical: (1,286,561rps) (avg: 777ns)
fromCanonicalTrusted: (1,625,362rps) (avg: 615ns)
fromRaw: (1,313,004rps) (avg: 761ns)
fromRawTrusted: (1,672,463rps) (avg: 597ns)
toCanonical: (6,103,543rps) (avg: 163ns)
toRaw: (6,235,448rps) (avg: 160ns)
Uuid6
=====
generate: (3,466,357rps) (avg: 288ns)
MIN: (5,244,292rps) (avg: 190ns)
MAX: (5,151,746rps) (avg: 194ns)
fromCanonical: (1,324,905rps) (avg: 754ns)
fromCanonicalTrusted: (1,676,541rps) (avg: 596ns)
fromRaw: (1,357,353rps) (avg: 736ns)
fromRawTrusted: (1,717,530rps) (avg: 582ns)
toCanonical: (5,061,822rps) (avg: 197ns)
toRaw: (4,839,125rps) (avg: 206ns)
UuidNil
=======
generate: (9,312,932rps) (avg: 107ns)
MIN: (5,158,703rps) (avg: 193ns)
MAX: (8,795,275rps) (avg: 113ns)
fromCanonical: (1,293,946rps) (avg: 772ns)
fromCanonicalTrusted: (1,629,605rps) (avg: 613ns)
fromRaw: (1,472,042rps) (avg: 679ns)
fromRawTrusted: (1,780,904rps) (avg: 561ns)
toCanonical: (5,169,323rps) (avg: 193ns)
toRaw: (5,196,170rps) (avg: 192ns)
Uuid processing Uuid1
=====================
generate: (4,159,340rps) (avg: 240ns)
MIN: (4,877,918rps) (avg: 205ns)
MAX: (4,907,348rps) (avg: 203ns)
fromCanonical: (1,045,214rps) (avg: 956ns)
fromCanonicalTrusted: (1,255,223rps) (avg: 796ns)
fromRaw: (1,021,436rps) (avg: 979ns)
fromRawTrusted: (1,268,213rps) (avg: 788ns)
Uuid processing Uuid4
=====================
generate: (5,695,823rps) (avg: 175ns)
MIN: (4,886,337rps) (avg: 204ns)
MAX: (4,907,325rps) (avg: 203ns)
fromCanonical: (1,047,372rps) (avg: 954ns)
fromCanonicalTrusted: (1,292,729rps) (avg: 773ns)
fromRaw: (1,031,590rps) (avg: 969ns)
fromRawTrusted: (1,266,122rps) (avg: 789ns)
Uuid processing Uuid6
=====================
generate: (4,122,279rps) (avg: 242ns)
MIN: (4,744,102rps) (avg: 210ns)
MAX: (4,860,271rps) (avg: 205ns)
fromCanonical: (1,066,004rps) (avg: 938ns)
fromCanonicalTrusted: (1,298,925rps) (avg: 769ns)
fromRaw: (1,053,871rps) (avg: 948ns)
fromRawTrusted: (1,286,373rps) (avg: 777ns)
Uuid processing UuidNil
=======================
generate: (8,140,742rps) (avg: 122ns)
MIN: (4,717,779rps) (avg: 211ns)
MAX: (8,261,012rps) (avg: 121ns)
fromCanonical: (1,052,765rps) (avg: 949ns)
fromCanonicalTrusted: (1,285,968rps) (avg: 777ns)
fromRaw: (1,130,468rps) (avg: 884ns)
fromRawTrusted: (1,312,878rps) (avg: 761ns)
Much of this library would not exist without the great work and documentation of other engineers. In particular:
- ksuid: an in-depth exploration of the guid nuances
- ulid: an elegant solution to a persistent problem
- uuid-random: allocating pages of randomness is by far the biggest performance factor
Also, thank you:
- ruleb: researching and patching worker support
Feel free to make a branch and send a pull request through github
Please report any issues or bugs through github