auto-increment id for revisions about digdag HOT 3 CLOSED

Implementation idea:

• digdag push command generates a random if -r option is not set
  • discussion: another idea is to use local timestamp (OffsetDateTime e.g. 2016-06-01T01:02:03+0900). But I think this is unnecessary because revisions.created_at is saved anyways.
  • discussion: another idea is to use SHA1 hash of project archive. But this is not good because users can't upload the same contents twice as 2 different revisions.

Future idea:

• Create a REST API and/or CLI command that returns sequential index of revisions in a project. With it, a revision can be referenced by projName#index format (e.g. myproj#1, myproj#2, ...).
  • But this causes a potential performance problem: calculating index of the revisions need O(N) CPU time on RDBMS (select row_number() over (partition by project_id order by id)). If an user pushed a lot of revisions to a projects, listing up revisions may take time.
  • To solve this issue, revisions table need index column. But setting that number at INSERT is not obvious implementation on RDBMS.
  • And I don't think such index is important to have. So, this is TODO.

from digdag.

Agree, using a UUID if a -r is not specified makes sense for now.

from digdag.

Already resolved. Close it.

from digdag.