techscientist / ksuid

K-Sortable Globally Unique IDs

ksuid

ksuid is a Go library that can generate and parse KSUIDs.

What is a KSUID?

KSUID is for K-Sortable Unique IDentifier. It's a way to generate globally unique IDs similar to RFC 4122 UUIDs, but contain a time component so they can be "roughly" sorted by time of creation. The remainder of the KSUID is randomly generated bytes.

Why use KSUIDs?

Distributed systems often require unique IDs. There are numerous solutions out there for doing this, so why KSUID?

1. Sortable by Timestamp

Unlike the more common choice of UUIDv4, KSUIDs contain a timestamp component that allows them to be roughly sorted by generation time. This is obviously not a strong guarantee as it depends on wall clocks, but is still incredibly useful in practice.

2. No Coordination Required

Snowflake IDs[1] and derivatives require coordination, which significantly increases the complexity of implementation and creates operations overhead. While RFC 4122 UUIDv1s do have a time component, there aren't enough bytes of randomness to provide strong protections against duplicate ID generation.

KSUIDs use 128-bits of pseudorandom data, which provides a 64-times larger number space than the 122-bits in the well-accepted RFC 4122 UUIDv4 standard. The additional timestamp component drives down the extremely rare chance of duplication to the point of near physical infeasibility, even assuming extreme clock skew (> 24-hours) that would cause other severe anomalies.

1. https://blog.twitter.com/2010/announcing-snowflake

3. Lexographically Sortable, Portable Representations

The binary and string representations are lexicographically sortable, which allows them to be dropped into systems which do not natively support KSUIDs and retain their k-sortable characteristics.

The string representation is that it is base62-encoded, so that they can "fit" anywhere alphanumeric strings are accepted.

How do they work?

KSUIDs are 20-bytes: a 32-bit unsigned integer UTC timestamp and a 128-bit randomly generated payload. The timestamp uses big-endian encoding, to allow lexicographic sorting. The timestamp epoch is adjusted to March 5th, 2014, providing over 100 years of useful life starting at UNIX epoch + 14e8. The payload uses a cryptographically-strong pseudorandom number generator.

The string representation is fixed at 27-characters encoded using a base62 encoding that also sorts lexicographically.

Command Line Tool

This package comes with a simple command-line tool ksuid. This tool can generate KSUIDs as well as inspect the internal components for debugging purposes.