IDs sequenced by time? about uuid HOT 21 CLOSED

RFC 4122 specifies an explicit ordering of the fields - http://tools.ietf.org/html/rfc4122#section-4.1.2 - that puts the low bytes of the timestamp at the front of the string. I forget the rational for this - something to do with putting the most-likely-to-change fields at the front of the string to improve string comparison performance, iirc. If you want to sort by timestamp, you'll need to swap the 1st and 3rd fields (separated by '-'), then order lexicographically.

from uuid.

> var id = '110ec58a-a0f2-4ac4-8393-c866d813b8d1';
undefined
> id.replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1')
"4ac4-a0f2-110ec58a-8393-c866d813b8d1"

Dealing with non-standard id formats is a slippery slope to which there is no bottom. And since this can pretty easily be bolted on outside of this module I don't see any particular reason to include it here. Sorry. :-/

from uuid.

@sm2017: Good question.

I edited my last comment to remove the four field form (the one that resulted in clockseq being the primary sort field). Readers should just stick to the replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1') form.

While an argument can be made for including clockseq in the sort order because it gets incremented if there's a system clock regression, the fact it is randomly initialized on process startup makes it unsuitable for inclusion in sorting. (Including it in my previous comment was a mistake on my part. My apologies for the confusion.)

What this means is that in the (very rare?) case where the system clock regresses (goes back in time), the sort order won't reflect the order in which ids were generated. I expect that's a minor concern for most folks, however.

(Note: The behavior of clockseq described here is part of the 4122 spec, where the intent is to minimize the risk of uuid collisions.)

from uuid.

Only for v1 (timestamped) uuids. You'd have to lexicographically order by time_hi, time_mid, time_lo , and clock_sequence fields, respectively.

const uuidV1 = require('uuid/v1');

function v1UuidCompare(a, b) {
  a = a.replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1');
  b = b.replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1');
  return a < b ? -1 : (a > b ? 1 : 0);
}

> v1UuidCompare(uuidV1(), uuidV1())
-1

Edit: Remove clockseq as primary sort field, since it's non-deterministic across process restarts.

from uuid.

So I think it's possible to transform a UUID v1 to an ordered UUID v1 and back like this:

Just so we're clear, there is no provision in RFC4122 for "ordered" v1 UUIDs. Describing them as such is a misnomer. (In fact, the transformation described creates an explicitly invalid UUID, because it moves the version field as well as the time_hi field)

Better to just say "transform a v1 UUID to a sort-friendly string"

from uuid.

I see. That's interesting and should be documented I think. How would you swap these two fields best in Javascript?

Or better, why don't you introduce a new option for that. Or introduce a new, unofficial version uuid.vX() for that?

from uuid.

:(

But you said, the RFC 4122 says to put the low bytes of the timestamp at the front??

PS: thx for the code snippet

from uuid.

Sorry, by "low bytes" I meant "low fields". The fields are lowest-first but within a field the hex octets are ordered highest-first. So the regex above is what you want. (But test it to make sure!)

From the RFC:

Each field is treated as an integer and has its value printed as a zero-filled hexadecimal digit string with the most significant digit first. The hexadecimal values "a" through "f" are output as lower case characters and are case insensitive on input.

from uuid.

I seeee ... thx. Yeah, I already tested and your regex seems to be correct. Cheers!

from uuid.

Hello guys,

uuid.v4().replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1') < uuid.v4().replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1')

doesn't always result in true.
what am I missing here ?

from uuid.

@xerosanyam, v4 uuids are randomly generated.

from uuid.

So there is no way to sequence IDs by time ? 🤔

from uuid.

@broofa Why you have 2 logic?

replace(/^(.{8})-(.{4})-(.{4})-(.{4})/, \'$4-$3-$2-$1\')

and

replace(/^(.{8})-(.{4})-(.{4})/, '$3-$2-$1')

from uuid.

@broofa Thanks a lot

from uuid.

In RFC 4122 UUID v1 format, the first three number groups aaaaaaaa-bbbb-cccc are generated from the low, middle, and high parts of a timestamp.

So I think it's possible to transform a UUID v1 to an ordered UUID v1 and back like this:

const uuidV1 = require('uuid/v1');

function orderUUIDv1(regularUuid) {
	return regularUuid.replace(/^(.{4})(.{4})-(.{4})-(.{4})/, '$4$3-$1-$2');
}

function unorderUUIDv1(orderedUuid) {
	return orderedUuid.replace(/^(.{4})(.{4})-(.{4})-(.{4})/, '$3$4-$2-$1');
}

// a9ec5a07-e872-11e9-b92c-080027d0eccd
let regularUuid = uuidV1();

// 11e9e872-a9ec-5a07-b92c-080027d0eccd
let orderedUuid = orderUUIDv1(regularUuid);

// a9ec5a07-e872-11e9-b92c-080027d0eccd
let unorderedUuid = unorderUUIDv1(orderedUuid);

This works by swapping the time-low and time-high parts (the first and third groups of hexadecimal digits, respectively). Basically moving the more rapidly varying part to the right.

This is inspired by how MySQL 8 deals with UUIDs. Specifically UUID_TO_BIN and BIN_TO_UUID, and explained here.

This is also possible in PHP with Ramsey\Uuid using OrderedTimeCodec.

from uuid.

@broofa Yes, the result is an invalid UUID string, but it can be used to store it in an optimized way in relational databases, specially MySQL.

Main disadvantages of storing regular UUIDs are:

1. UUID is not ordered, so inserts are random and the data is scattered.
   • Fixed by transforming it to a 'sort-friendly string'.
2. UUID has 36 characters which make it bulky. When used as a PRIMARY KEY it makes the index bigger which sometimes cannot fit into memory.
   • Fixed by storing it in BINARY.

The fact that the MySQL added this functionality in MySQL 8 is a good indicator that it is useful.

Of course there is no need to do it JavaScript, you can transform to ordered UUIDs in MySQL or PHP, but I post it here in case it is helpful to anyone.

from uuid.

FYI I built this awhile back which does this shuffling when converting uuidv1 into a binary format: https://github.com/odo-network/binary-uuid

from uuid.

Hi, is there any ways to generate uuid in descending order?

from uuid.

What exactly are you trying to achieve, @rajatkeshar ?

from uuid.

I think he means a version 1 UUID with the time fields swapped so the lowest bytes appear first, the middle bytes are next, and the highest bytes come last so that the UUID is sorted by creation time.

Like in Ramsey Ordered-Time UUID or MySQL 8 Ordered UUID.

Or maybe the experimental UUID version 6?

from uuid.

But generating these would still result in ascending order…

from uuid.