#unique-identifier #sortable #uuid #identifier #ulid #unique-id #primary-key

small_uid

A 64 bit lexicographically sortable unique identifier implementation

1 unstable release

new 0.2.2 Nov 17, 2024
0.2.1 Oct 24, 2024
0.1.2 Oct 19, 2024

#56 in Database implementations

Download history 336/week @ 2024-10-15 190/week @ 2024-10-22 60/week @ 2024-10-29 35/week @ 2024-11-05 126/week @ 2024-11-12

432 downloads per month

MIT/Apache

14KB
209 lines

Small UID

Crates.io License (version) Crates.io Total Downloads

⚠️ This project is in experimental phase, the API may may be subject to change.

This is a Rust implementation of Small UIDs, generating unique identifiers with short strings and lexicographically sortable. Check that repository for implementation in other languages.

UUIDs are frequently used as database Primary Key in software development. However, they aren't the best choice mainly due to their random sorting and the resulting fragmentation in databases indexes.

Using ULIDs is generally a very good alternative, solving most of UUID flaws.

Twitter's Snowflake is another option if you want to generate roughly sortable uid. But, Snowflake is not using random numbers instead it used machine id to generate the uid. It's a good choice if you integrate it into a distributed systems and doesn't really need randomness.

Small UIDs are also an ideal alternative when you do not need as much uniqueness and want shorter "user-friendly" encoded strings.

Introduction

Small UIDs are short unique identifiers especially designed to be used as efficient database Primary Key:

  • Half smaller than UUID / ULID (64-bit)
  • Lexicographically sortable
  • Encodable as a short user-friendly and URL-safe base-64 string (a-zA-Z0-9_-)
  • User-friendly strings are generated in a way to be always very different (no shared prefix due to similar timestamps)
Small UID ULID UUID v4
Size 64 bits 128 bits 128 bits
Monotonic sort order Yes *** Yes No
Random bits 20 80 122
Collision odds ** 1,024 / ms* 1.099e+12 / ms* 2.305e+18

* theorical number of generated uids before the first expected collision.
** the uid includes a timestamp, so collisions may occur only during the same millisecond.
*** monotonic sort order, but random order when generated at the same millisecond.

They are internally stored as 64-bit integers (44-bit timestamp followed by 20 random bits):

|-----------------------|  |------------|
        Timestamp            Randomness
         44 bits               20 bits

The random number suffix still guarantees a decent amount of uniqueness when many ids are created in the same millisecond (up to 1,048,576 different values) and you may only expect collision if you're generating more than 1024 random ids during the same millisecond.

Sorting

Because of the sequential timestamp, Small UIDs are naturally sorted chronologically. It improves indexing when inserting values in databases, new ids being appended to the end of the table without reshuffling existing data (read more in this article).

However, sort order within the same millisecond is not guaranteed because of the random bits suffix.

Examples of usage

Generating Small UIDs

let smalluid1 = SmallUid::new();
let smalluid2 = SmallUid::try_from("GSntNvOw6n8".to_string()).unwrap();

Converting Small UIDs

let smalluid = SmallUid::new();
let uid_string = smalluid.to_string();

Dependencies

~1.2–1.8MB
~27K SLoC