How to create a GUID / UUID

I'm trying to create globally-unique identifiers in JavaScript. I'm not sure what routines are available on all browsers, how "random" and seeded the built-in random number generator is, etc.

The GUID / UUID should be at least 32 characters and should stay in the ASCII range to avoid trouble when passing them around.

[Edited 2021-10-16 to reflect latest best-practices for producing RFC4122-complaint UUIDs]

Most readers here will want to use the uuid module. It is well-tested and supported.

The crypto.randomUUID() function is an emerging standard that is supported in Node.js and an increasing number of browsers.

If neither of those work for you, there is this method (based on the original answer to this question):

function uuidv4() {
  return ([1e7]+-1e3+-4e3+-8e3+-1e11).replace(/[018]/g, c =>
    (c ^ crypto.getRandomValues(new Uint8Array(1))[0] & 15 >> c / 4).toString(16)
  );
}

console.log(uuidv4());

Note: The use of any UUID generator that relies on Math.random() is strongly discouraged (including snippets featured in previous versions of this answer) for reasons best-explained here. TL;DR: Math.random()-based solutions do not provide good uniqueness guarantees.

UUIDs (Universally Unique IDentifier), also known as GUIDs (Globally Unique IDentifier), according to RFC 4122, are identifiers designed to provide certain uniqueness guarantees.

While it is possible to implement RFC-compliant UUIDs in a few lines of JavaScript code (e.g., see @broofa's answer, below) there are several common pitfalls:

• Invalid id format (UUIDs must be of the form "xxxxxxxx-xxxx-Mxxx-Nxxx-xxxxxxxxxxxx", where x is one of [0-9, a-f] M is one of [1-5], and N is [8, 9, a, or b]
• Use of a low-quality source of randomness (such as Math.random)

Thus, developers writing code for production environments are encouraged to use a rigorous, well-maintained implementation such as the uuid module.

I really like how clean Broofa's answer is, but it's unfortunate that poor implementations of Math.random leave the chance for collision.

Here's a similar RFC4122 version 4 compliant solution that solves that issue by offsetting the first 13 hex numbers by a hex portion of the timestamp, and once depleted offsets by a hex portion of the microseconds since pageload. That way, even if Math.random is on the same seed, both clients would have to generate the UUID the exact same number of microseconds since pageload (if high-perfomance time is supported) AND at the exact same millisecond (or 10,000+ years later) to get the same UUID:

function generateUUID() { // Public Domain/MIT
    var d = new Date().getTime();//Timestamp
    var d2 = ((typeof performance !== 'undefined') && performance.now && (performance.now()*1000)) || 0;//Time in microseconds since page-load or 0 if unsupported
    return 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, function(c) {
        var r = Math.random() * 16;//random number between 0 and 16
        if(d > 0){//Use timestamp until depleted
            r = (d + r)%16 | 0;
            d = Math.floor(d/16);
        } else {//Use microseconds since page-load if supported
            r = (d2 + r)%16 | 0;
            d2 = Math.floor(d2/16);
        }
        return (c === 'x' ? r : (r & 0x3 | 0x8)).toString(16);
    });
}

var onClick = function(){
    document.getElementById('uuid').textContent = generateUUID();
}
onClick();

#uuid { font-family: monospace; font-size: 1.5em; }

<p id="uuid"></p>
<button id="generateUUID" onclick="onClick();">Generate UUID</button>

Here's a fiddle to test.

Modernized snippet for ES6

const generateUUID = () => {
  let
    d = new Date().getTime(),
    d2 = (performance && performance.now && (performance.now() * 1000)) || 0;
  return 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, c => {
    let r = Math.random() * 16;
    if (d > 0) {
      r = (d + r) % 16 | 0;
      d = Math.floor(d / 16);
    } else {
      r = (d2 + r) % 16 | 0;
      d2 = Math.floor(d2 / 16);
    }
    return (c == 'x' ? r : (r & 0x7 | 0x8)).toString(16);
  });
};

const onClick = (e) => document.getElementById('uuid').textContent = generateUUID();

document.getElementById('generateUUID').addEventListener('click', onClick);

onClick();

#uuid { font-family: monospace; font-size: 1.5em; }

<p id="uuid"></p>
<button id="generateUUID">Generate UUID</button>

broofa's answer is pretty slick, indeed - impressively clever, really... RFC4122 compliant, somewhat readable, and compact. Awesome!

But if you're looking at that regular expression, those many replace() callbacks, toString()'s and Math.random() function calls (where he's only using four bits of the result and wasting the rest), you may start to wonder about performance. Indeed, joelpt even decided to toss out an RFC for generic GUID speed with generateQuickGUID.

But, can we get speed and RFC compliance? I say, YES! Can we maintain readability? Well... Not really, but it's easy if you follow along.

But first, my results, compared to broofa, guid (the accepted answer), and the non-rfc-compliant generateQuickGuid:

                  Desktop   Android
           broofa: 1617ms   12869ms
               e1:  636ms    5778ms
               e2:  606ms    4754ms
               e3:  364ms    3003ms
               e4:  329ms    2015ms
               e5:  147ms    1156ms
               e6:  146ms    1035ms
               e7:  105ms     726ms
             guid:  962ms   10762ms
generateQuickGuid:  292ms    2961ms
  - Note: 500k iterations, results will vary by browser/CPU.

So by my 6th iteration of optimizations, I beat the most popular answer by over 12 times, the accepted answer by over 9 times, and the fast-non-compliant answer by 2-3 times. And I'm still RFC 4122 compliant.

Interested in how? I've put the full source on http://jsfiddle.net/jcward/7hyaC/3/ and on http://jsperf.com/uuid-generator-opt/4

For an explanation, let's start with broofa's code:

function broofa() {
    return 'xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'.replace(/[xy]/g, function(c) {
        var r = Math.random()*16|0, v = c == 'x' ? r : (r&0x3|0x8);
        return v.toString(16);
    });
}

console.log(broofa())

So it replaces x with any random hexadecimal digit, y with random data (except forcing the top two bits to 10 per the RFC spec), and the regex doesn't match the - or 4 characters, so he doesn't have to deal with them. Very, very slick.

The first thing to know is that function calls are expensive, as are regular expressions (though he only uses 1, it has 32 callbacks, one for each match, and in each of the 32 callbacks it calls Math.random() and v.toString(16)).

The first step toward performance is to eliminate the RegEx and its callback functions and use a simple loop instead. This means we have to deal with the - and 4 characters whereas broofa did not. Also, note that we can use String Array indexing to keep his slick String template architecture:

function e1() {
    var u='',i=0;
    while(i++<36) {
        var c='xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx'[i-1],r=Math.random()*16|0,v=c=='x'?r:(r&0x3|0x8);
        u+=(c=='-'||c=='4')?c:v.toString(16)
    }
    return u;
}

console.log(e1())

Basically, the same inner logic, except we check for - or 4, and using a while loop (instead of replace() callbacks) gets us an almost 3X improvement!

The next step is a small one on the desktop but makes a decent difference on mobile. Let's make fewer Math.random() calls and utilize all those random bits instead of throwing 87% of them away with a random buffer that gets shifted out each iteration. Let's also move that template definition out of the loop, just in case it helps:

function e2() {
    var u='',m='xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx',i=0,rb=Math.random()*0xffffffff|0;
    while(i++<36) {
        var c=m[i-1],r=rb&0xf,v=c=='x'?r:(r&0x3|0x8);
        u+=(c=='-'||c=='4')?c:v.toString(16);rb=i%8==0?Math.random()*0xffffffff|0:rb>>4
    }
    return u
}

console.log(e2())

This saves us 10-30% depending on platform. Not bad. But the next big step gets rid of the toString function calls altogether with an optimization classic - the look-up table. A simple 16-element lookup table will perform the job of toString(16) in much less time:

function e3() {
    var h='0123456789abcdef';
    var k='xxxxxxxx-xxxx-4xxx-yxxx-xxxxxxxxxxxx';
    /* same as e4() below */
}
function e4() {
    var h=['0','1','2','3','4','5','6','7','8','9','a','b','c','d','e','f'];
    var k=['x','x','x','x','x','x','x','x','-','x','x','x','x','-','4','x','x','x','-','y','x','x','x','-','x','x','x','x','x','x','x','x','x','x','x','x'];
    var u='',i=0,rb=Math.random()*0xffffffff|0;
    while(i++<36) {
        var c=k[i-1],r=rb&0xf,v=c=='x'?r:(r&0x3|0x8);
        u+=(c=='-'||c=='4')?c:h[v];rb=i%8==0?Math.random()*0xffffffff|0:rb>>4
    }
    return u
}

console.log(e4())

The next optimization is another classic. Since we're only handling four bits of output in each loop iteration, let's cut the number of loops in half and process eight bits in each iteration. This is tricky since we still have to handle the RFC compliant bit positions, but it's not too hard. We then have to make a larger lookup table (16x16, or 256) to store 0x00 - 0xFF, and we build it only once, outside the e5() function.

var lut = []; for (var i=0; i<256; i++) { lut[i] = (i<16?'0':'')+(i).toString(16); }
function e5() {
    var k=['x','x','x','x','-','x','x','-','4','x','-','y','x','-','x','x','x','x','x','x'];
    var u='',i=0,rb=Math.random()*0xffffffff|0;
    while(i++<20) {
        var c=k[i-1],r=rb&0xff,v=c=='x'?r:(c=='y'?(r&0x3f|0x80):(r&0xf|0x40));
        u+=(c=='-')?c:lut[v];rb=i%4==0?Math.random()*0xffffffff|0:rb>>8
    }
    return u
}

console.log(e5())

I tried an e6() that processes 16-bits at a time, still using the 256-element LUT, and it showed the diminishing returns of optimization. Though it had fewer iterations, the inner logic was complicated by the increased processing, and it performed the same on desktop, and only ~10% faster on mobile.

The final optimization technique to apply - unroll the loop. Since we're looping a fixed number of times, we can technically write this all out by hand. I tried this once with a single random variable, r, that I kept reassigning, and performance tanked. But with four variables assigned random data up front, then using the lookup table, and applying the proper RFC bits, this version smokes them all:

var lut = []; for (var i=0; i<256; i++) { lut[i] = (i<16?'0':'')+(i).toString(16); }
function e7()
{
    var d0 = Math.random()*0xffffffff|0;
    var d1 = Math.random()*0xffffffff|0;
    var d2 = Math.random()*0xffffffff|0;
    var d3 = Math.random()*0xffffffff|0;
    return lut[d0&0xff]+lut[d0>>8&0xff]+lut[d0>>16&0xff]+lut[d0>>24&0xff]+'-'+
    lut[d1&0xff]+lut[d1>>8&0xff]+'-'+lut[d1>>16&0x0f|0x40]+lut[d1>>24&0xff]+'-'+
    lut[d2&0x3f|0x80]+lut[d2>>8&0xff]+'-'+lut[d2>>16&0xff]+lut[d2>>24&0xff]+
    lut[d3&0xff]+lut[d3>>8&0xff]+lut[d3>>16&0xff]+lut[d3>>24&0xff];
}

console.log(e7())

Modualized: http://jcward.com/UUID.js - UUID.generate()

The funny thing is, generating 16 bytes of random data is the easy part. The whole trick is expressing it in string format with RFC compliance, and it's most tightly accomplished with 16 bytes of random data, an unrolled loop and lookup table.

I hope my logic is correct -- it's very easy to make a mistake in this kind of tedious bit work. But the outputs look good to me. I hope you enjoyed this mad ride through code optimization!

Be advised: my primary goal was to show and teach potential optimization strategies. Other answers cover important topics such as collisions and truly random numbers, which are important for generating good UUIDs.

Use:

let uniqueId = Date.now().toString(36) + Math.random().toString(36).substring(2);

document.getElementById("unique").innerHTML =
  Math.random().toString(36).substring(2) + (new Date()).getTime().toString(36);

<div id="unique">
</div>

If IDs are generated more than 1 millisecond apart, they are 100% unique.

If two IDs are generated at shorter intervals, and assuming that the random method is truly random, this would generate IDs that are 99.99999999999999% likely to be globally unique (collision in 1 of 10^15).

You can increase this number by adding more digits, but to generate 100% unique IDs you will need to use a global counter.

If you need RFC compatibility, this formatting will pass as a valid version 4 GUID:

let u = Date.now().toString(16) + Math.random().toString(16) + '0'.repeat(16);
let guid = [u.substr(0,8), u.substr(8,4), '4000-8' + u.substr(13,3), u.substr(16,12)].join('-');

let u = Date.now().toString(16)+Math.random().toString(16)+'0'.repeat(16);
let guid = [u.substr(0,8), u.substr(8,4), '4000-8' + u.substr(13,3), u.substr(16,12)].join('-');
document.getElementById("unique").innerHTML = guid;

<div id="unique">
</div>

The above code follow the intention, but not the letter of the RFC. Among other discrepancies it's a few random digits short. (Add more random digits if you need it) The upside is that this is really fast :) You can test validity of your GUID here