feat(cloud-scheduler): renew active leases

2026-07-12 17:23:07 +08:00
parent b0a5b1426f
commit 1b15a8a218
4 changed files with 225 additions and 2 deletions
@@ -577,3 +577,175 @@ def test_expired_assignment_cannot_be_claimed(database_url: str) -> None:
        assert task.status == "assigned"
    finally:
        database.close()
+
+
+def test_active_lease_renews_for_owning_host(database_url: str) -> None:
+    database = CloudDatabase(database_url)
+    host_id = _unique_id("renew-host")
+    device_id = _unique_id("renew-device")
+    task_id = _unique_id("renew-task")
+    now = datetime(2026, 7, 12, 8, 0, tzinfo=UTC)
+    initial_expiry = now + timedelta(minutes=1)
+    renewed_expiry = now + timedelta(minutes=2)
+
+    try:
+        database.repository.upsert_host(host_id, address=None, last_seen_at=now)
+        database.repository.replace_host_devices(
+            host_id,
+            [_device(device_id, host_id)],
+        )
+        database.repository.enqueue_task(
+            ScheduledTask(
+                id=task_id,
+                goal="renew me",
+                workflow_definition_id=None,
+                constraints=TaskConstraints(),
+                created_at=now,
+            )
+        )
+        database.repository.assign_task(
+            task_id=task_id,
+            host_id=host_id,
+            device_id=device_id,
+            lease_id="renew-lease",
+            lease_expires_at=initial_expiry,
+            now=now,
+        )
+        database.repository.claim_assignment(host_id=host_id, now=now)
+
+        status = database.repository.renew_lease(
+            task_id=task_id,
+            attempt=1,
+            lease_id="renew-lease",
+            host_id=host_id,
+            lease_expires_at=renewed_expiry,
+            now=now + timedelta(seconds=30),
+        )
+
+        assert status == "renewed"
+        task = database.repository.get_task(task_id)
+        assert task is not None
+        assert task.lease_expires_at == renewed_expiry
+        attempts = database.repository.list_task_attempts(task_id)
+        assert attempts[0].lease_expires_at == renewed_expiry
+    finally:
+        database.close()
+
+
+@pytest.mark.parametrize(
+    ("attempt", "lease_id", "host_id"),
+    [
+        (2, "lease-current", "owner"),
+        (1, "lease-stale", "owner"),
+        (1, "lease-current", "foreign"),
+    ],
+)
+def test_stale_or_foreign_lease_renewal_conflicts(
+    database_url: str,
+    attempt: int,
+    lease_id: str,
+    host_id: str,
+) -> None:
+    database = CloudDatabase(database_url)
+    owner_id = _unique_id("renew-owner")
+    device_id = _unique_id("renew-conflict-device")
+    task_id = _unique_id("renew-conflict-task")
+    now = datetime(2026, 7, 12, 9, 0, tzinfo=UTC)
+    initial_expiry = now + timedelta(minutes=1)
+
+    try:
+        database.repository.upsert_host(owner_id, address=None, last_seen_at=now)
+        database.repository.replace_host_devices(
+            owner_id,
+            [_device(device_id, owner_id)],
+        )
+        database.repository.enqueue_task(
+            ScheduledTask(
+                id=task_id,
+                goal="do not renew",
+                workflow_definition_id=None,
+                constraints=TaskConstraints(),
+                created_at=now,
+            )
+        )
+        database.repository.assign_task(
+            task_id=task_id,
+            host_id=owner_id,
+            device_id=device_id,
+            lease_id="lease-current",
+            lease_expires_at=initial_expiry,
+            now=now,
+        )
+
+        status = database.repository.renew_lease(
+            task_id=task_id,
+            attempt=attempt,
+            lease_id=lease_id,
+            host_id=owner_id if host_id == "owner" else _unique_id("foreign"),
+            lease_expires_at=now + timedelta(minutes=2),
+            now=now + timedelta(seconds=30),
+        )
+
+        assert status == "conflict"
+        task = database.repository.get_task(task_id)
+        assert task is not None
+        assert task.lease_expires_at == initial_expiry
+    finally:
+        database.close()
+
+
+def test_expired_or_missing_lease_cannot_be_renewed(database_url: str) -> None:
+    database = CloudDatabase(database_url)
+    host_id = _unique_id("expired-renew-host")
+    device_id = _unique_id("expired-renew-device")
+    task_id = _unique_id("expired-renew-task")
+    now = datetime(2026, 7, 12, 10, 0, tzinfo=UTC)
+
+    try:
+        database.repository.upsert_host(host_id, address=None, last_seen_at=now)
+        database.repository.replace_host_devices(
+            host_id,
+            [_device(device_id, host_id)],
+        )
+        database.repository.enqueue_task(
+            ScheduledTask(
+                id=task_id,
+                goal="expired renewal",
+                workflow_definition_id=None,
+                constraints=TaskConstraints(),
+                created_at=now,
+            )
+        )
+        database.repository.assign_task(
+            task_id=task_id,
+            host_id=host_id,
+            device_id=device_id,
+            lease_id="expired-renew-lease",
+            lease_expires_at=now + timedelta(seconds=1),
+            now=now,
+        )
+
+        assert (
+            database.repository.renew_lease(
+                task_id=task_id,
+                attempt=1,
+                lease_id="expired-renew-lease",
+                host_id=host_id,
+                lease_expires_at=now + timedelta(minutes=2),
+                now=now + timedelta(seconds=2),
+            )
+            == "expired"
+        )
+        assert (
+            database.repository.renew_lease(
+                task_id=_unique_id("missing-task"),
+                attempt=1,
+                lease_id="missing-lease",
+                host_id=host_id,
+                lease_expires_at=now + timedelta(minutes=2),
+                now=now,
+            )
+            == "not_found"
+        )
+    finally:
+        database.close()