Combine Stripe IDs with Snowflake IDs and you get... pika! - the last ID system you'll ever need, combining pragmatism with functionality.
Example ID: user_MTI5Njg4Njg1MDQwODg5ODYx
- Object type prefixes
- Type prefix atomicity
- Guaranteed multi-node space-time uniqueness
- Timestamped
- Fast & simple
- Shorter than UUIDs
- Standalone
- Option to be cryptographically secure
- Unable to sequence by integer (pikas are strings)
- Slower generation (by a few nanoseconds - pika is 1.5m ops/sec vs snowflake 2m ops/sec on an M1, however all Snowflakes and Pikas have a theoretical limit of generating a maximum of 1,024,000 IDs per node per second, so there won't be a difference in throughput either way)
- Slightly larger sizes (pikas are ~28 bytes vs Snowflake's 8 bytes)
Pika IDs consist of 2 sections: the type prefix and the tail, separated by an underscore.
When creating a pika ID, you must specify the prefix to be prepended - the general rule of thumb should be to use a different prefix for each object type (e.g. user
, team
, post
, etc).
Type prefixes should be lowercase, short, alphanumeric strings. If you have an object type with a long name, then it's recommended to shorten it down into an acronym or similar. For example, if we had an object type called "connected account", then we'd make the type prefix ca
- or even if we had a type called "channel", we might want to shorten it down to ch
- it's up to you to decide what you think makes the most distinctive sense.
The tail is the part that comes after the underscore (e.g. MTI5Njg4Njg1MDQwODg5ODYx
). Usually, this is just a base64-encoded Snowflake ID, however, if the pika is cryptographically secure, then the base64 decoded string value will start with an s_
prefix, followed by a cryptographically random string, then followed by another underscore and the Snowflake ID.
Example of a normal decoded tail:
129688685040889861
Example of a cryptographically secure decoded tail:
s_387d0775128c383fa8fbf5fd9863b84aba216bcc6872a877_129688685040889861
To guarantee that developers use the correct pre-defined prefix types for the right object types, pika requires you to "register" them before they're used to prevent warnings from being thrown. This is also where you define if a prefix type should be cryptographically secure or not.