Worker coordination with file locks

src/alpine_bits_python/api.py

@@ -45,6 +45,7 @@ from .rate_limit import (
     webhook_limiter,
 )
 from .reservation_service import ReservationService
+from .worker_coordination import is_primary_worker
 
 # Configure logging - will be reconfigured during lifespan with actual config
 _LOGGER = get_logger(__name__)
@@ -182,24 +183,16 @@ async def push_listener(customer: DBCustomer, reservation: DBReservation, hotel)
 async def lifespan(app: FastAPI):
     # Setup DB
 
-    # Determine if this is the primary worker
+    # Determine if this is the primary worker using file-based locking
     # Only primary runs schedulers/background tasks
     # In multi-worker setups, only one worker should run singleton services
-    worker_id = os.environ.get("APP_WORKER_ID", "0")
-    is_primary_worker = worker_id == "0"
-
-    # For uvicorn with --workers, detect if we're the main process
-    if not is_primary_worker:
-        # Check if running under uvicorn's supervisor
-        is_primary_worker = (
-            multiprocessing.current_process().name == "MainProcess"
-        )
+    is_primary, worker_lock = is_primary_worker()
 
     _LOGGER.info(
         "Worker startup: process=%s, pid=%d, primary=%s",
         multiprocessing.current_process().name,
         os.getpid(),
-        is_primary_worker,
+        is_primary,
     )
 
     try:
@@ -217,9 +210,9 @@ async def lifespan(app: FastAPI):
     # Setup logging from config with email monitoring
     # Only primary worker should have the report scheduler running
     email_handler, report_scheduler = setup_logging(
-        config, email_service, loop, enable_scheduler=is_primary_worker
+        config, email_service, loop, enable_scheduler=is_primary
     )
-    _LOGGER.info("Application startup initiated (primary_worker=%s)", is_primary_worker)
+    _LOGGER.info("Application startup initiated (primary_worker=%s)", is_primary)
 
     DATABASE_URL = get_database_url(config)
     engine = create_async_engine(DATABASE_URL, echo=False)
@@ -260,7 +253,7 @@ async def lifespan(app: FastAPI):
     _LOGGER.info("Database tables checked/created at startup.")
 
     # Hash any existing customers (only in primary worker to avoid race conditions)
-    if is_primary_worker:
+    if is_primary:
         async with AsyncSessionLocal() as session:
             customer_service = CustomerService(session)
             hashed_count = await customer_service.hash_existing_customers()
@@ -311,6 +304,10 @@ async def lifespan(app: FastAPI):
     await engine.dispose()
     _LOGGER.info("Application shutdown complete")
 
+    # Release worker lock if this was the primary worker
+    if worker_lock:
+        worker_lock.release()
+
 
 async def get_async_session(request: Request):
     async_sessionmaker = request.app.state.async_sessionmaker