feat(cloud-scheduler): reap expired leases

This commit is contained in:
2026-07-12 17:32:11 +08:00
parent e1af4403f5
commit c58d9e5316
3 changed files with 242 additions and 1 deletions
@@ -451,6 +451,59 @@ class SQLAlchemyCloudRepository:
attempt_row.result_json = result_json
return "recorded"
def reap_expired_leases(
self,
*,
now: datetime,
max_attempts: int,
) -> list[str]:
if max_attempts < 1:
raise ValueError("max_attempts must be at least 1")
with self._sessions.begin() as session:
statement = (
select(ScheduledTaskRow)
.where(
ScheduledTaskRow.status.in_(("assigned", "dispatched")),
ScheduledTaskRow.lease_expires_at.is_not(None),
ScheduledTaskRow.lease_expires_at <= _iso(now),
)
.order_by(ScheduledTaskRow.lease_expires_at, ScheduledTaskRow.id)
)
if self.engine.dialect.name == "postgresql":
statement = statement.with_for_update(skip_locked=True)
tasks = session.scalars(statement).all()
reaped_task_ids: list[str] = []
for task in tasks:
attempt_row = session.get(
TaskAttemptRow,
(task.id, task.attempt_count),
with_for_update=self.engine.dialect.name == "postgresql",
)
if attempt_row is None:
continue
attempt_row.status = "expired"
attempt_row.completed_at = _iso(now)
attempt_row.failure_reason = "lease expired"
task.updated_at = _iso(now)
task.lease_id = None
task.lease_expires_at = None
task.result_json = None
if task.attempt_count < max_attempts:
task.status = "queued"
task.assigned_host_id = None
task.assigned_device_id = None
task.failure_reason = None
else:
task.status = "failed"
task.failure_reason = (
f"lease expired after {task.attempt_count} attempts"
)
reaped_task_ids.append(task.id)
return reaped_task_ids
def list_task_attempts(self, task_id: str) -> list[Any]:
with self._sessions() as session:
rows = session.scalars(