[MDEV-23748] support not-MAC-address based UUID versions Created: 2020-09-17  Updated: 2023-10-31

Status: Open
Project: MariaDB Server
Component/s: Server
Fix Version/s: None

Type: Task Priority: Minor
Reporter: Jonathan Shapiro Assignee: Unassigned
Resolution: Unresolved Votes: 1
Labels: beginner-friendly

Issue Links:
PartOf
includes MDEV-11339 Support UUID v4 generation Open
includes MDEV-32637 Implement native UUID7 function Open
Relates
relates to MDEV-4958 Adding datatype UUID Closed

 Description   

So far as I know, this affects all versions of MariaDB.

On Linux, FreeBSD, and (perhaps) MacOS, MariaDB uses the host MAC address as the basis for generating v1 UUID values in the UUID() function. On Windows, it generates a random nonce and uses that instead of a MAC address.

In cloud or virtual machine deployments, it is likely that

  1. MAC addresses are not unique (e.g. all Hyper-V installations use the same small pool by default), or
  2. Multiple copies of MariaDB may end up running in adjacent containers that execute on the same underlying machine (and therefore share a MAC address).

On Windows deployments this has not been an issue, because these deployments generate random nonces instead of using the MAC address.

It would be helpful if it were possible to configure the Linux/FreeBSD/?Mac installations to use such a nonce in environments where the uniqueness of the MAC address is uncertain or unreliable. Since the nonce support code is already present, one might hope this would not be a big lift.

 Comments   
Comment by Vladislav Vaintroub [ 2020-09-17 ]

shap, "On Windows, it generates a random nonce and uses that instead of a MAC address."
Really? Maybe I'm missing something , but my_gethwaddr is implemented there, which serves as uuid suffix.

Comment by Sergei Golubchik [ 2020-09-18 ]

v1 UUID must use the mac address.

So this is, basically, a (valid) request to implement UUID version 3, 4, or 5.

Comment by Trevor Gross [ 2022-01-21 ]

Keeping UUID() to be a UUIDv1 generator based on mac address seems necessary for MySQL compatibility.

As a random user, I'd vouch for implementing UUID generation functions the way that the postgres plugin does it https://www.postgresql.org/docs/13/uuid-ossp.html

  • uuid_generate_v1() : UUIDv1 generator using real HW mac address (same as UUID() is currently for My/Maria)
  • uuid_generate_v1mc(): UUIDv1 but use a randomly generated mac. It makes sense to me that this random mac should be saved as a global variable, for consistency if the DB is restarted or moved (unsure how PG handles this).
  • uuid_generate_v3(namespace, name): UUIDv3, MD5 namespace
  • uuid_generate_v4(): UUIDv4, full random UUID
  • uuid_generate_v5(namespace, name): UUIDv5, SHA-1 namespace

In my opinion, uuid_generate_v4 and uuid_generate_v1mc would be the most useful to have. The mc version with the mac saved to a variable would nicely address circumstances where generating node needs to be tracked (so v4 isn't usable), but tracked to the database instance rather than the server it runs on.

Generated at Thu Feb 08 09:24:44 UTC 2024 using Jira 8.20.16#820016-sha1:9d11dbea5f4be3d4cc21f03a88dd11d8c8687422.