[LoopedBard3]

Summary

A late or repeated DELETE can change retained job metadata from Deleted back to Deleting after its run group has finished. The worker selects new groups through New jobs, so the old run need not be processed again even though cleanup has already completed.

• Make DELETE a state-preserving no-op for already-deleted jobs while retaining existing 202/Location and missing-ID 204 responses.
• Synchronize its guarded mutation with the worker's terminal publication using the same per-job monitor. Capture the previous state under that monitor and log afterward, with no cleanup or await inside the critical section.
• Add focused regressions for late/repeated requests, all current non-Deleted states, missing IDs, and both orderings of the real production transitions. A small internal worker transition helper avoids reflection-heavy worker setup in tests.

Verification

Reproduced the defect against freshly fetched upstream main at 0d35cad5102892b091d9a88a52ea1f7a91bb1381. The original terminal-preservation assertions failed with expected Deleted, actual Deleting.

dotnet test .\test\Microsoft.Crank.UnitTests\Microsoft.Crank.UnitTests.csproj --no-restore --filter "FullyQualifiedName~JobDeletionTests|FullyQualifiedName~JobsControllerTests"

Final result on Windows/.NET 10: 39 passed, including 16 new cases and 23 existing controller tests. The race tests also reject a guard moved outside the controller lock and an unlocked worker publication; the correct implementation was restored and revalidated afterward.

This verifies the current-upstream defect and local fix, not the source revision of any previously deployed binary. No benchmark, deployment, Linux, or live-agent validation is claimed. Other state-writer races and controller behavior are outside this change.

[Copilot]

🟢 Approval recommended

The shared monitor correctly serializes the two transitions, existing responses are preserved, and no unresolved issues were identified.

No open issues 🎉

What changed in this PR

Prevents late or repeated DELETE requests from moving completed jobs back from Deleted to Deleting, while preserving existing HTTP responses.

Changes:

• Guards DELETE state changes with the same per-job monitor used for worker completion.
• Extracts terminal-state publication into an internal helper, keeping cleanup and logging outside the lock.
• Adds regression tests for repeated requests, missing jobs, non-deleted states, and both concurrent transition orderings.

File	Description
test/​Microsoft.Crank.UnitTests/​JobDeletionTests.cs	Tests deletion behavior and synchronized terminal-state preservation.
src/​Microsoft.Crank.Agent/​Startup.cs	Publishes Deleted under the job monitor and logs afterward.
src/​Microsoft.Crank.Agent/​Controllers/​JobsController.cs	Preserves Deleted under the shared monitor without changing responses.

_{Review effort: Balanced}