Files
kandev/docs/plans/stale-queued-prompt-count/task-02-session-delete-cancels-queue.md
T
2026-08-13 16:00:15 +01:00

2.7 KiB

title, status, parallelism, parallel-safe
title status parallelism parallel-safe
Cancel session queue on session delete pending sequential no

Task 02: Cancel Session Queue On Session Delete

Goal

Deleting a session removes its pending queue entries and publishes message.queue.status_changed for the owning task so the sidebar badge no longer counts orphan rows.

Root Cause Slice

orchestrator.Service.DeleteSession deletes the session row and returns. It never calls messageQueue.CancelAll (or equivalent) and never publishes queue status. CountPendingByTaskIDs still counts those rows by task_id.

RED Regression (must fail before the fix)

In apps/backend/internal/orchestrator/ (prefer beside existing session / queue tests):

  1. Create task T with sessions S1 and S2.
  2. Enqueue N pending prompts on S1 (and optionally some on S2).
  3. Call DeleteSession(S1) successfully.
  4. Assert:
    • messageQueue has no pending entries for S1;
    • CountPendingByTask(T) equals only S2's remaining pending count;
    • a message.queue.status_changed event was published with task_id = T (and preferably session_id = S1).

Name suggestion: TestDeleteSessionCancelsQueuedPromptsAndPublishesStatus.

Confirm RED: S1 rows remain and/or no status event / task count unchanged.

Implementation

In DeleteSession, after the session row is successfully deleted (or in the same critical section after the delete is durable — prefer after successful DB delete so a failed delete does not empty the queue):

  1. messageQueue.CancelAll(ctx, sessionID) (or a dedicated purge-by-session if CancelAll's reserved-in-flight semantics are wrong for teardown — match "session is gone, nothing should remain pending for it").
  2. Publish queue status for the task using the helper from Task 01 (task_id + session_id).
  3. Best-effort log on cancel/publish failure; do not fail the session delete after the row is already gone (document that choice in the test).

If CancelAll preserves reserved-in-flight durable lifecycle rows, add an explicit purge-by-session for deletion teardown so orphans cannot linger.

Acceptance

  • RED then GREEN on the regression test.
  • Task badge / CountPendingByTask no longer includes deleted-session rows.
  • Deleting one session does not wipe another session's queue on the same task.
  • Task 01's purge publish helper is reused rather than a third publish path.

Validation

cd apps/backend
go test ./internal/orchestrator/ -count=1

Dependencies

Task 01 (publish helper / event shape).

Output Contract

Report RED/GREEN evidence, whether CancelAll or a stronger purge was needed for reserved rows, changed files, and plan/task status updates.