Universally Unique Prefixed Lexicographically Sortable Identifier
Project description
UPID
pronounced YOO-pid
aka Universally Unique Prefixed Lexicographically Sortable Identifier
This is the spec and Python implementation for UPID.
UPID is based on ULID but with some modifications, inspired by this article and Stripe IDs.
The core idea is that a meaningful prefix is specified that is stored in a 128-bit UUID-shaped slot. Thus a UPID is human-readable (like a Stripe ID), but still efficient to store, sort and index.
UPID allows a prefix of up to 4 characters (will be right-padded if shorter than 4), includes a non-wrapping timestamp with about 300 millisecond precision, and 64 bits of entropy.
This is a UPID in Python:
upid("user") # user_aaccvpp5guht4dts56je5a
And in Rust:
UPID::new("user") // user_aaccvpp5guht4dts56je5a
Specification
Key changes relative to ULID:
- Uses a modified form of Crockford's base32 that uses lower-case and includes the full alphabet (for prefix flexibility).
- Does not permit upper-case/lower-case to be decoded interchangeably.
- The text encoding is still 5 bits per base32 character.
- 20 bits assigned to the prefix
- 40 bits (down from 48) assigned to the timestamp, placed first in binary for sorting
- 64 bits (down from 80) for randomness
- 4 bits as a version specifier
user aaccvpp5 guht4dts56je5 a
|----| |--------| |-------------| |-----|
prefix time random version total
4 chars 8 chars 13 chars 1 char 26 chars
\________/________________|___________ |
/ | \ |
/ | \ |
40 bits 64 bits 24 bits 128 bits
5 bytes 8 bytes 3 bytes 16 bytes
time random prefix+version
Binary layout
0 1 2 3
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| time_high |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| time_low | random |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| random |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
| random | prefix_and_version |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
Collision
Relative to ULID, the time precision is reduced from 48 to 40 bits (keeping the most significant bits, so oveflow still won't occur until 10889 AD), and the randomness reduced from 80 to 64 bits.
The timestamp precision at 40 bits is around 300 milliseconds. In order to have a 50% probability of collision with 64 bits of randomness, you would need to generate around 4 billion items per 100 millisecond window.
Python implementation
This aims to be maximally simple to convey the core working of the spec. The current Python implementation is entirely based on mdomke/python-ulid.
Rust implementation
The current Rust implementation is based on dylanhart/ulid-rs, but using the same lookup base32 lookup method as the Python implementation.
Development
rye sync
rye run all # or fmt/lint/check/test
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.