Harden queue worker crashes and POST validation

This commit is contained in:
Dymas
2026-05-16 11:14:52 +02:00
parent 92762b07d5
commit d5f81d15b7
7 changed files with 150 additions and 59 deletions
+78 -52
View File
@@ -768,6 +768,29 @@ class DownloadQueue:
self._save_job_locked(job)
self._cleanup_staging_dir(job)
def _fail_job_runtime(self, job, exc, process=None):
if process is not None:
try:
os.killpg(os.getpgid(process.pid), signal.SIGTERM)
except OSError:
try:
process.terminate()
except OSError:
pass
message = f"Download worker failed unexpectedly: {exc}"
with self.lock:
job["status"] = "failed"
job["exit_code"] = 1
job["pid"] = None
job["finished_at"] = now_iso()
job["updated_at"] = job["finished_at"]
job.setdefault("log", []).append(message)
self.current_process = None
self.current_job_id = None
self.current_job = None
self._save_job_locked(job)
self._cleanup_staging_dir(job)
def _run(self):
while not self.stop_event.is_set():
job = self._next_pending()
@@ -830,56 +853,59 @@ class DownloadQueue:
job["pid"] = process.pid
self._save_job_locked(job)
assert process.stdout is not None
for line in process.stdout:
self._append_log(job, line)
exit_code = process.wait()
finalize_error = None
moved_files = []
watchlist_sync_error = None
cleanup_staging = False
if exit_code == 0 and not job.get("cancel_requested"):
try:
moved_files = finalize_library_files(job)
except Exception as exc:
finalize_error = exc
else:
if self.watchlist_sync_fn is not None:
try:
self.watchlist_sync_fn(job)
except Exception as exc:
watchlist_sync_error = exc
try:
assert process.stdout is not None
for line in process.stdout:
self._append_log(job, line)
exit_code = process.wait()
finalize_error = None
moved_files = []
watchlist_sync_error = None
cleanup_staging = False
if exit_code == 0 and not job.get("cancel_requested"):
try:
moved_files = finalize_library_files(job)
except Exception as exc:
finalize_error = exc
else:
if self.watchlist_sync_fn is not None:
try:
self.watchlist_sync_fn(job)
except Exception as exc:
watchlist_sync_error = exc
with self.lock:
canceled = job.get("cancel_requested")
job["exit_code"] = 1 if finalize_error else exit_code
job["pid"] = None
job["finished_at"] = now_iso()
job["updated_at"] = job["finished_at"]
if canceled:
job["status"] = "canceled"
job.setdefault("log", []).append("Canceled.")
elif finalize_error:
job["status"] = "failed"
job.setdefault("log", []).append(f"Library rename failed: {finalize_error}")
elif exit_code == 0:
job["status"] = "done"
for path in moved_files:
job.setdefault("log", []).append(f"Saved: {path}")
job.setdefault("log", []).append("Download completed.")
if watchlist_sync_error:
job.setdefault("log", []).append(f"Watchlist sync failed: {watchlist_sync_error}")
elif self.watchlist_sync_fn is not None:
job.setdefault("log", []).append("Watchlist download state synced.")
else:
job["status"] = "failed"
job.setdefault("log", []).append(f"Download failed with exit code {exit_code}.")
cleanup_staging = True
if canceled:
cleanup_staging = True
self.current_process = None
self.current_job_id = None
self.current_job = None
self._save_job_locked(job)
if cleanup_staging:
self._cleanup_staging_dir(job)
with self.lock:
canceled = job.get("cancel_requested")
job["exit_code"] = 1 if finalize_error else exit_code
job["pid"] = None
job["finished_at"] = now_iso()
job["updated_at"] = job["finished_at"]
if canceled:
job["status"] = "canceled"
job.setdefault("log", []).append("Canceled.")
elif finalize_error:
job["status"] = "failed"
job.setdefault("log", []).append(f"Library rename failed: {finalize_error}")
elif exit_code == 0:
job["status"] = "done"
for path in moved_files:
job.setdefault("log", []).append(f"Saved: {path}")
job.setdefault("log", []).append("Download completed.")
if watchlist_sync_error:
job.setdefault("log", []).append(f"Watchlist sync failed: {watchlist_sync_error}")
elif self.watchlist_sync_fn is not None:
job.setdefault("log", []).append("Watchlist download state synced.")
else:
job["status"] = "failed"
job.setdefault("log", []).append(f"Download failed with exit code {exit_code}.")
cleanup_staging = True
if canceled:
cleanup_staging = True
self.current_process = None
self.current_job_id = None
self.current_job = None
self._save_job_locked(job)
if cleanup_staging:
self._cleanup_staging_dir(job)
except Exception as exc:
self._fail_job_runtime(job, exc, process=process)