Storing a UUID in Cloud Spanner

Question

I would like to use a UUID as a primary key in Cloud Spanner. What is the best way to read and write UUIDs? Is there a UUID type, or client library support?

Answer 1

The simplest solution is just to store it as a STRING in the standard RFC 4122 format. E.g.:

"d1a0ce61-b9dd-4169-96a8-d0d7789b61d9"

This will take 37 bytes to store (36 bytes plus a length byte). If you really want to save every possible byte, you could store your UUID as two INT64's. However, you would need to write your own libraries for serializing/deserializing the values, and they wouldn't appear very pretty in your SQL queries. In most cases, the extra ~21 bytes of savings per row is probably not worth it.

Note that some UUID generation algorithms generate the UUID sequentially based on a timestamp. If the UUID values generated by a machine are monotonically increasing, then this can lead to hot-spotting in Cloud Spanner (this is analogous to the anti-pattern of using timestamps as the beginning of a primary key), so it is best to avoid these variants (e.g. UUID version 1 is not recommended).

This Stackoverflow answer provides more details about the various UUID versions. (TL;DR: use Version 4 with Cloud Spanner since a psuedo-ranndom number is used in the generation)

Answer 2

As per Cloud Spanner documentation:

There are several ways to store the UUID as the primary key:

• In a STRING(36) column.
• In a pair of INT64 columns.
• In a BYTES(16) column.