XCUID

eliXir CUID.

Collision-resistant ids optimized for horizontal scaling and binary search lookup performance.

This project is a more modern and optimized implementation compare to duailibe/cuid which is no longer maintained AFAIK.

Usage

Add xcuid to your list of dependencies in mix.exs:

def deps do
  [
    {:xcuid, "~> 0.1.0"}
  ]
end

Run mix deps.get, then generate a CUID is simple:

XCUID.generate() # => cka3hpvst29dw1m12hzwblxd5

If you want to start multiple generator you can do

{:ok, pid} = XCUID.start(name: :generator_1)
XCUID.generate(pid) #=> cka3hpvst29dw1m12hzwblxd5

Why not UUID v4?

Refer to the original project for the original motivation of CUID. Personally I use this instead of UUID v4 because:

• CUID is monotonically increasing (due to the timestamp) where UUID v4 (because it is completely random) is not. In addition to the benefits mentioned in the original project, monotonically increasing ID can be used to implement cursor based pagination and UUID can not.
• CUID is easier to add prefix compare to UUID, for example acct_cka3hpvst29dw1m12hzwblxd5. This is especially the case when used with Ecto and Postgres. In this case the DB column will be of type UUID (basically hexdecimal digits) so you can't store the prefix with the UUID. You will have to use custom Ecto.Type to do the conversion which is a pain. You can argue that you could just store the prefix and the UUID as string, but in that case it is much longer than a CUID and take more space to store. 32 characters (UUID) vs 25 character (CUID).
• Can be used as html ids any anywhere where things must start with a letter, where as UUID can't because it can start with number.
• Another minor thing is that the dashes in UUID makes selecting UUID not as smooth as CUID, because in a lot of environments for example in the browser or terminal when you double click to select a UUID it won't highlight the entire thing because dash are word breaks. Obviously you can remove the dashes when displaying or use some front-end magic, but that is extra work where CUID just works out of the box.

One disadvantage of CUID is that it take more space, UUID is 16 bytes if stored as hexdecimal, where as CUID is at least 25 bytes. However you can could save some space by not storing an extra inserted_at column because that is already included in CUID. If we subtract that 8 byte for timestamp then technically it is just 1 byte more than UUID which is a reasonable price to pay for getting the above advantages.

Performance

I have optimized the generation code to the best of my knowledge. Currently it can generate ~250k CUIDs per second on my MBP (Intel(R) Core(TM) i9-9880H CPU @ 2.30GHz). If you want even better performance you can pre-generate them in batches.

Breakdown

c - h72gsb32 - 0000 - udoc - l363eofy

The groups, in order, are:

• c identifies this as a cuid, and allows you to use it in html entity ids. This will be changed to d on May 25 2059. See the Others section for detail.
• h72gsb32 is the timestamp in millisecond
• 0000 is a counter. A single process might generate the same random string. The weaker the pseudo-random source, the higher the probability. That problem gets worse as processors get faster. So this counter is an attempt to solve the problem. The counter will roll over if the value gets too big. (i.e from zzzz back to 0000).
• udoc is a fingerprint, the first two characters are based on the OS process ID and the next two are based on the hostname.
• l363eofy a random string generated using :rand.uniform with a cryptographically strong seed.

Others

This issue in the original javascript version is fixed in XCUID by changing all CUID to start with d instead of c after the roll over time which is 2_821_109_907_456 in unix timestamp.

Credits

• Eric Elliott (author of original JavaScript version)
• Lucas Duailibe (author of the original elixir cuid)
• Roy Bao