fix(npu-workers): await _save_workers_to_config in _load_workers_from_config (#12526)

[mrveiss]

Thinking Path

_load_workers_from_config (sync) called the async _save_workers_to_config without await on the config-missing branch — the coroutine was created and immediately garbage-collected (silent no-op + RuntimeWarning: coroutine ... was never awaited). It ran via asyncio.to_thread(self._load_workers_from_config) from _initialize_impl, i.e. on a worker thread with no running event loop, so simply adding await in place wouldn't have worked — the whole call chain needed to become async so it runs on the event loop where await is meaningful.

Checked all other self._<name>( call sites against the async-method set in the file: the other 4 _save_workers_to_config() calls (add_worker, update_worker, remove_worker, update_load_balancing_config) already correctly await it. The only other bare-looking calls (_health_check_loop, failover_monitor, _pulse_probe_loop) are wrapped in asyncio.create_task(...) (intentional background tasks) and _get_worker_status inside asyncio.gather(...) (correctly awaited by gather) — no other bugs found in this file.

What Changed

• autobot-backend/services/npu_worker_manager.py:
  • _load_workers_from_config is now async def and directly await self._save_workers_to_config() on the missing-file branch.
  • _initialize_impl now does await self._load_workers_from_config() directly instead of asyncio.to_thread(self._load_workers_from_config).
  • Blocking file I/O (Path.exists(), open() + yaml.safe_load()) is still kept off the event loop via asyncio.to_thread, extracted into a small _read_config_file helper — same pattern _save_workers_to_config already uses for mkdir.
• autobot-backend/services/test_npu_worker_config_load_12526.py (new): regression tests covering
  • missing config file → save is awaited (mocked) and, separately, an end-to-end run where the file is actually written (the real regression check — previously this file was never created).
  • existing config file → save is not invoked.
  • _initialize_impl awaits _load_workers_from_config.

Verification

$ python3 -W error::RuntimeWarning -m pytest services/test_npu_worker_config_load_12526.py \
    services/test_npu_worker_events.py services/npu_worker_manager_pulse_test.py -p no:xdist -q
28 passed in 0.44s

No RuntimeWarning: coroutine ... was never awaited (elevated to a hard error via -W error::RuntimeWarning and the run still passed).

$ python3 -m black --check autobot-backend/services/npu_worker_manager.py autobot-backend/services/test_npu_worker_config_load_12526.py
All done! 2 files would be left unchanged.
$ python3 -m flake8 --max-line-length=120 autobot-backend/services/npu_worker_manager.py autobot-backend/services/test_npu_worker_config_load_12526.py
(no output — clean)

Closes #12526

Model Used

Claude Sonnet 5

[github-actions]

✅ SSOT Configuration Compliance: Passing

🎉 No hardcoded values detected that have SSOT config equivalents!