Add a manual purge action for completed Workflows

[daviddavis]

Summary

Add a manual API for purging completed Workflow records, mirroring the existing pulpcore POST /pulp/api/v3/tasks/purge/ action.

Motivation

Workflows accumulate the same way tasks do — once completed, the rows stay around for audit/history but eventually become noise. Pulpcore already provides this for tasks via the tasks_purge action; workflows should have the equivalent so operators have a way to keep the workflow table tidy.

Explicitly manual, not automatic: like pulpcore's task purge, the action should only run when an operator (or a scheduled job they configure themselves) invokes it. No background sweeping should be added.

How pulpcore does it (for reference)

• POST /pulp/api/v3/tasks/purge/ is an @action(detail=False, methods=["post"]) on TaskViewSet (pulpcore/app/viewsets/task.py).
• The action accepts a PurgeSerializer body with two fields (pulpcore/app/serializers/purge.py):
  • finished_before — datetime; defaults to 30 days ago.
  • states — set of allowed final states; defaults to ["completed"], restricted to TASK_FINAL_STATES.
• The action dispatches pulpcore.app.tasks.purge.purge and returns an OperationPostponedResponse (202).
• The purge task is itself a regular pulp task: it filters in chunks of 1000, scopes by current domain and core.delete_task permission, and emits ProgressReports for totals, per-type counts, and errors.
• A tasks entry is registered in INITIAL_GUARDED_SCHEDULES (pulpcore/app/util.py) so a default schedule exists, but the user is in control of whether it actually runs.

Proposed design for pulp_workflow

Add a purge action on WorkflowViewSet:

• POST /pulp/api/v3/workflows/purge/ (action, detail=False, methods=["post"]).
• Request body — WorkflowPurgeSerializer:
  • finished_before — DateTimeField, default now() - 30d. Only workflows whose finished_at is strictly less than this are eligible.
  • states — MultipleChoiceField over the workflow's terminal states (COMPLETED, FAILED, CANCELED); default ["completed"].
• Dispatch a new task pulp_workflow.app.tasks.purge_workflows and return OperationPostponedResponse (202). Do not run synchronously.
• Permission model:
  • Add a delete_workflow Django permission to Workflow.Meta.permissions (or rely on the default delete permission once added) and gate the action on it via DEFAULT_ACCESS_POLICY, mirroring pulpcore's pattern of filtering by delete_task inside the purge task.
  • Scope by current domain (pulp_domain=get_domain()) when the task was dispatched by a user (i.e. not from a schedule).
• Inside purge_workflows:
  • Build a queryset of Workflow.objects.filter(finished_at__lt=finished_before, state__in=states).
  • Delete in chunks of 1000 to keep memory bounded (same DELETE_LIMIT pattern as pulpcore).
  • Emit ProgressReports with codes such as purge.workflows.total, purge.workflows.key.<model>, purge.workflows.error for parity with the task purge.
• Cascade behavior: WorkflowTask, WorkflowTaskArg, and WorkflowTaskKwarg already cascade off Workflow via on_delete=CASCADE, so deleting the Workflow cleans those up automatically. The associated TaskGroup (once Associate Workflows with a pulpcore TaskGroup #3 lands) and the child Task rows are not owned by the workflow and should be left alone — they are governed by tasks_purge. Document this in the action's help text so users understand they need to run both.

Explicitly out of scope

• No automatic/scheduled purging by default. Do not add a TaskSchedule entry that runs on its own. Operators who want recurring purges can create their own TaskSchedule pointing at pulp_workflow.app.tasks.purge_workflows.
• No purge of running or waiting workflows. The states field must reject non-terminal states (validate in the serializer the same way pulpcore's PurgeSerializer restricts to TASK_FINAL_STATES).
• No deletion of the Task rows the workflow dispatched. Those are pulpcore's responsibility.

Acceptance criteria

• POST /pulp/api/v3/workflows/purge/ exists, accepts finished_before and states, returns 202 with a task reference.
• Dispatched task removes only workflows whose finished_at < finished_before and whose state is in the (validated terminal) states list.
• Workflows in non-terminal states are never deleted.
• Cascading delete of WorkflowTask / WorkflowTaskArg / WorkflowTaskKwarg works.
• Default finished_before is "30 days ago" and default states is ["completed"].
• Domain scoping matches pulpcore's task purge (current-domain only when dispatched by a user).
• Permission check prevents users without delete_workflow from purging.
• Functional test covers: defaults spare recent workflows; explicit finished_before in the future deletes everything terminal; non-terminal workflows are spared regardless.

Related

• pulpcore reference implementation: pulpcore/app/tasks/purge.py, pulpcore/app/viewsets/task.py, pulpcore/app/serializers/purge.py.