leefer/xiaobaifupan
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

BAI-3: harden live readiness diagnostics

Browse Source 
Co-authored-by: multica-agent <github@multica.ai>
This commit is contained in:
MS-01-Codex
2026-08-07 12:29:01 +08:00
co-authored by multica-agent
parent a326df2a8d
commit 777af0cb8b
8 changed files with 308 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files
app/backend/jobs/runner.py
+12 -6
View File
@@ -51,8 +51,7 @@ class InProcessJobRunner:
lock = self._lock(definition.lock_key)
if not lock.acquire(blocking=False):
return False
self._execute_locked(job_id, idempotency_key, action, metadata, lock)
return True
return self._execute_locked(job_id, idempotency_key, action, metadata, lock)
def start_scheduler(
self, callback: Callable[[], None], interval_seconds: float,
@@ -105,13 +104,19 @@ class InProcessJobRunner:
def _execute_locked(
self, job_id: str, idempotency_key: str, action: JobAction,
metadata: dict[str, Any] | None, lock: threading.Lock,
) -> None:
) -> bool:
definition = self.registry.get(job_id)
try:
for attempt in range(1, definition.max_attempts + 1):
run_id = self.repository.start(
job_id, idempotency_key, definition.output_version, metadata
job_id,
idempotency_key,
definition.output_version,
metadata,
definition.timeout_seconds,
)
if run_id is None:
return False
started = time.perf_counter()
try:
result = action()
@@ -119,7 +124,7 @@ class InProcessJobRunner:
raise RuntimeError(str(result.get("error") or "Job reported failure"))
elapsed_ms = round((time.perf_counter() - started) * 1000)
self.repository.finish(run_id, "success", elapsed_ms)
return
return True
except Exception as exc:
elapsed_ms = round((time.perf_counter() - started) * 1000)
self.repository.finish(
@@ -127,7 +132,8 @@ class InProcessJobRunner:
type(exc).__name__, str(exc),
)
if attempt >= definition.max_attempts:
return
return True
return True
finally:
lock.release()