Current Series Release Notes

19.0.0-10

New Features

  • Registering a client returns the uuid of the record along with its client_id.

  • GET and DELETE /v2/{project_id}/clients/{uuid} accept a client uuid as well as a client_id.

  • The command freezer-manage db purge has been added. It permanently deletes soft deleted rows. It accepts optional --age-in-days DAYS parameter to specify the age of the rows to delete. Default value is 30 days. Supported by the SQLAlchemy driver only.

Upgrade Notes

  • A client is identified by its project, client_id and hostname. Several schedulers can register under one client_id, each with its own record. Before, a second scheduler took over the record of the first one.

  • A central client_id is unique across all projects. A central registration is rejected if another project uses the client_id, and any registration is rejected if another project uses it as a central client.

  • A registration with a uuid that has no record is rejected as not found. Register without the uuid to get a new record.

  • GET and DELETE with a client_id shared by several clients are rejected as unprocessable. Use the uuid instead.

  • A job is now validated against every freezer-scheduler serving its client. Several schedulers may share a client_id and the coordination ring may hand the job to any of them, so the job is refused unless all of them support its actions. It was previously checked against an arbitrary one of them, so a job the chosen scheduler could run was accepted even when a sibling could not, and that sibling then skipped it without saying so. The rejection names the uuid of the scheduler that lacks the capability; one that has been decommissioned, and is narrowing what its client supports, is removed with DELETE /v2/{project_id}/clients/{uuid}.

Bug Fixes

  • Deleting non-existent resources (backups, jobs, actions, sessions, and clients) now returns HTTP 404 Not Found (DocumentNotFound) instead of HTTP 204 No Content, properly indicating when a resource cannot be deleted because it does not exist.

  • The ‘all_projects’ parameter in the jobs API was overwritten by the policy check, so admins always got jobs from all projects no matter what was in the request. The policy is now checked only when ‘all_projects’ is set to True. If the policy does not allow listing all projects, 403 Forbidden is returned.