Uuid v7 collisions

[vudaltsov]

Description

From time to time I get v7 collisions. I also noticed that since 0ba1ffb uuids of this version do not differ much in the random part, maybe, that's the reason?

Steps to reproduce

<?php

// generate_uuids_v7.php

use function Ramsey\Uuid\v7;

require_once __DIR__.'/vendor/autoload.php';

for ($j = 0; $j < 100; ++$j) {
    $uuids = '';

    for ($i = 0; $i < 1_000_000; ++$i) {
        $uuids .= v7()."\n";
    }

    file_put_contents('uuids_v7.txt', $uuids, $j === 0 ? 0 : FILE_APPEND);
}

php generate_uuids_v7.php && sort uuids_v7.txt | uniq -cd

Expected behavior

Empty output

Screenshots or output

   2 018b2847-0e47-7148-8dec-df216b0d76c7
   2 018b2847-a950-7367-9fa7-e3344774c890
   2 018b2847-c3b4-7203-be93-5976ad423647
   2 018b2847-fbbd-703e-b616-d256da1f0ec0
   2 018b2848-1f15-7200-b831-ddc7c0e6f86f
   2 018b2848-dcbf-72eb-b6ea-256c9e1f5d38

Environment details

• version of this package: 4.7.4
• PHP version: 8.2.10
• OS: macOS Ventura (13.3.1)

[vudaltsov]

Running same code with "ramsey/uuid": "4.5.1" does not give any collisions. So the bug is in new UnixTimeGenerator.

[ramsey]

The new code in UnixTimeGenerator was mostly borrowed from symfony/uid. Can you reproduce the issue using their method for generating v7 UUIDs?

[vudaltsov]

Can you reproduce the issue using their method for generating v7 UUIDs?

Yes, done.

[nicolas-grekas]

See symfony/symfony#52367 for the fix

[ramsey]

Thanks, @nicolas-grekas!

[vudaltsov]

I will do a PR.

[fannyfan414]

@ramsey @nicolas-grekas Hi! Have a question about uuid v7 collisions.

I need to replace 4-5 chars after generation uuid v7 with my specific characters (servers id or smth else).
At what positions would you advise me to change them so as not to greatly increase the chance of collisions?

[vudaltsov]

@fannyfan414 , you can use UUID v8, it allows to put arbitrary content.

[ramsey]

I would recommend using UUIDv8 for your use-case, by the way. It's intended for custom layouts like the one you're using.

To minimize chance of collision, I would probably place the server ID in the bytes to the far right of the UUID layout.