kandev/docs/plans/azure-devops-integration/task-24-watcher-persistence.md

id, title, status, wave, depends_on, plan, spec

id	title	status	wave	depends_on	plan	spec
24-watcher-persistence	Azure watcher persistence	completed	12	22-task-work-item-links	plan.md	../../specs/integrations/requirements/azure-devops-integration.md

Task 24: Azure Watcher Persistence

Acceptance

• Separate work-item and PR watcher stores implement workspace-scoped CRUD, enabled/due listing, error timestamps, deleting state, reset preview/reset, and cleanup-policy deletion. Common fields include workflow/step, Kandev repository/base branch, agent/executor profiles, prompt, poll interval, cleanup policy, optional in-flight cap, generation, last poll/error, and timestamps.
• Work-item filters persist Azure project ID and WIQL. PR filters persist Azure project ID, optional azure_repository_id, status, creator, and reviewer; the Azure filter is never confused with the Kandev repository_id.
• Reservation uniqueness is (watch, generation, project, work item) or (watch, generation, project, Azure repository, PR). Reserve, attach, and release all require the caller's generation; ownership loss is a terminal result, not a retry against the new generation.
• Controllers require workspace authorization for list/create and authorize the watch's owner before ID-based mutations without leaking another workspace's existence.
• Poll interval defaults to 300 seconds and clamps values below 60 seconds. max_inflight_tasks uses omitted/zero/positive semantics shared by other watches.
• GET /:id/reset/preview returns only tasks owned by the current watch generation. Reset increments generation before shared cleanup and removes prior reservations afterward. Delete marks the watch deleting and disabled before cleanup. Workspace deletion removes watches and reservations.

TDD Sequence

1. Add migration/store tests for both watch kinds, restart persistence, reservation uniqueness, generation ownership loss, and workspace cleanup; run them red before adding schema/store methods.
2. Add service/controller tests for validation, cross-workspace ID access, error recording, reset preview, cleanup policies, and deleting state; run them red before adding routes.
3. Implement schema, stores, reset service, and controllers. Refactor shared watch-kind helpers only after both suites are green.

Verification

• go test ./internal/azuredevops ./internal/orchestrator from apps/backend — passed (1,455 tests).
• Coverage includes workspace isolation, restart persistence, generation-aware reservations, reset preview/reset, deleting state, and workspace cleanup.

Files Likely Touched

• apps/backend/internal/azuredevops/watch_models.go
• apps/backend/internal/azuredevops/store_watches.go
• apps/backend/internal/azuredevops/store_watches_test.go
• apps/backend/internal/azuredevops/store_watch_reset.go
• apps/backend/internal/azuredevops/store_watch_reset_test.go
• apps/backend/internal/azuredevops/service_watches.go
• apps/backend/internal/azuredevops/service_watches_test.go
• apps/backend/internal/azuredevops/service_watch_reset.go
• apps/backend/internal/azuredevops/service_watch_reset_test.go
• apps/backend/internal/azuredevops/controller_watches.go
• apps/backend/internal/azuredevops/controller_watch_reset.go
• apps/backend/internal/azuredevops/controller_test.go
• apps/backend/internal/azuredevops/lifecycle.go

Dependencies

Task 22.

Parallelism

Sequential. Both watcher kinds share schema, reset, cleanup, and authorization.

Inputs

• Spec: Azure watches data model/API/state guarantees.
• GitLab watch models/store/reset as the closest two-kind provider pattern.
• internal/watchreset shared cleanup helper.

Risks

• Generation must be included in every reservation write/update condition so a reset cannot let an old dispatch attach a task to new ownership.
• ID-addressed routes must not trust a caller-supplied workspace ID before loading and authorizing the watch's persisted owner.

Output Contract

Report schema and state transitions, RED/GREEN commands, files changed, authorization evidence, risks, and update task/plan status.