sched: record resumed stopped results live and idempotently (B3)

2026-09-09 17:27:36 +07:00
parent ada0239a1c
commit b094385077
5 changed files with 357 additions and 3 deletions
@@ -384,3 +384,89 @@ def test_manual_and_other_schedule_runs_are_overlap_independent(
    assert result["a"].startswith("admit:run-")
    assert result["b"] == "exhausted"
    sched.ownership.release()
+
+
+def test_record_resumed_stopped_result_is_idempotent_and_attributed(
+    tmp_path: Path,
+) -> None:
+    """Live resumed results record exactly once with schedule attribution (B3)."""
+    from wf_api.run_lifecycle import persist_admission
+
+    from tests.scheduling.controlled import fixture_environment
+
+    sched, store, runs, sources = _harness(tmp_path, script={"*": "interrupt"})
+    t0 = ts(2026, 9, 8, 12, 0)
+    _add(sched, store, sources, "a", OneShotSource(t0), t0 - timedelta(seconds=1))
+    assert sched.poll(t0)["a"].startswith("admit:run-")
+    [run] = runs.list_runs()
+    assert run.status.value == "interrupted"
+    assert len([r for r in _history(store, "a") if r["kind"] == "interrupted"]) == 1
+
+    now = t0 + timedelta(minutes=1)
+    resumed_ckpt = f"{run.id}.000002"
+    assert (
+        sched.record_resumed_stopped_result(
+            run.id,
+            status_value="completed",
+            checkpoint_id=resumed_ckpt,
+            now=now,
+        )
+        is True
+    )
+    # Repeats (poll retries, restart recovery) never duplicate the entry.
+    assert (
+        sched.record_resumed_stopped_result(
+            run.id,
+            status_value="completed",
+            checkpoint_id=resumed_ckpt,
+            now=now,
+        )
+        is False
+    )
+    completed = [r for r in _history(store, "a") if r["kind"] == "completed"]
+    assert len(completed) == 1
+    assert completed[0]["run_id"] == run.id
+    assert completed[0]["checkpoint_id"] == resumed_ckpt
+    assert completed[0]["revision"] == 1
+    assert completed[0]["reason"] == "resumed"
+    # A resumed re-interruption is a new result (new checkpoint), not a dup.
+    assert (
+        sched.record_resumed_stopped_result(
+            run.id,
+            status_value="interrupted",
+            checkpoint_id=f"{run.id}.000003",
+            now=now,
+        )
+        is True
+    )
+    assert len([r for r in _history(store, "a") if r["kind"] == "interrupted"]) == 2
+    # Unknown runs, manual runs, and bad statuses record nothing.
+    assert (
+        sched.record_resumed_stopped_result(
+            "run-999999",
+            status_value="completed",
+            checkpoint_id=None,
+            now=now,
+        )
+        is False
+    )
+    assert (
+        sched.record_resumed_stopped_result(
+            run.id, status_value="bogus", checkpoint_id=None, now=now
+        )
+        is False
+    )
+    manual = persist_admission(
+        store=runs,
+        run_id=runs.allocate_run_id(),
+        environment=fixture_environment(object()),
+        resolved_input={},
+        max_steps=None,
+    )
+    assert (
+        sched.record_resumed_stopped_result(
+            manual.id, status_value="completed", checkpoint_id=None, now=now
+        )
+        is False
+    )
+    sched.ownership.release()