fix: fail loudly on misconfigured DATABASE_URL or REDIS_URL at startup

app/adapters/http/app.py

@@ -41,12 +41,14 @@ async def lifespan(app: FastAPI):
     redis_url = os.environ.get("REDIS_URL", "").strip()
     if redis_url:
         try:
-            from app.infra.queue.queue import create_queue
-            queue = await create_queue(redis_url)
-            if queue is not None:
-                _deps.get_async_service()._queue = queue
-        except Exception:
-            pass  # arq not installed or Redis unreachable — fall back silently
+            from arq import create_pool
+            from arq.connections import RedisSettings
+            pool = await create_pool(RedisSettings.from_dsn(redis_url))
+            from app.infra.queue.queue import ArqJobQueue
+            _deps.get_async_service()._queue = ArqJobQueue(pool)
+        except Exception as exc:
+            logger.error("REDIS_URL is set but Redis is unreachable: %s", exc)
+            raise RuntimeError(f"REDIS_URL is set but Redis is unreachable: {exc}") from exc
 
     yield
 

app/adapters/http/deps.py

@@ -30,16 +30,13 @@ async def init_job_store() -> JobStore:
 
     db_url = os.environ.get("DATABASE_URL", "").strip()
     if db_url:
+        from app.infra.jobs.postgres_job_store import PostgresJobStore
         try:
-            from app.infra.jobs.postgres_job_store import PostgresJobStore
             _job_store = await PostgresJobStore.create_pool(dsn=db_url)
-            return _job_store
         except Exception as exc:
-            _log.error(
-                "Failed to connect to Postgres (%s). "
-                "Falling back to SQLite — THIS IS NOT SAFE IN PRODUCTION.",
-                exc,
-            )
+            _log.error("DATABASE_URL is set but Postgres is unreachable: %s", exc)
+            raise RuntimeError(f"DATABASE_URL is set but Postgres is unreachable: {exc}") from exc
+        return _job_store
 
     from app.infra.jobs.sqlite_job_store import SQLiteJobStore
     _job_store = SQLiteJobStore()

app/infra/queue/worker.py

@@ -4,9 +4,12 @@
 Run the worker:
     arq app.infra.queue.worker.WorkerSettings
 """
+import logging
 import os
 from typing import Any
 
+logger = logging.getLogger(__name__)
+
 
 async def run_inference(ctx: dict, job_id: str, model: str, version: str, payload: Any) -> None:
     from uuid import UUID
@@ -61,7 +64,11 @@ async def startup(ctx: dict) -> None:
     db_url = os.environ.get("DATABASE_URL", "").strip()
     if db_url:
         from app.infra.jobs.postgres_job_store import PostgresJobStore
-        store = await PostgresJobStore.create_pool(dsn=db_url)
+        try:
+            store = await PostgresJobStore.create_pool(dsn=db_url)
+        except Exception as exc:
+            logger.error("DATABASE_URL is set but Postgres is unreachable: %s", exc)
+            raise RuntimeError(f"DATABASE_URL is set but Postgres is unreachable: {exc}") from exc
     else:
         from app.infra.jobs.sqlite_job_store import SQLiteJobStore
         store = SQLiteJobStore()

docs/guides/production-deployment.md

@@ -19,6 +19,8 @@ REDIS_URL=redis://localhost:6379/0
 
 When `ENV=production` and `API_KEYS` is not set, the server refuses to start.
 
+When `DATABASE_URL` or `REDIS_URL` is set but the service is unreachable, the server (and arq worker) refuse to start with a clear error instead of silently falling back to SQLite or in-process async. Ensure both services are healthy before starting the application.
+
 ---
 
 ## Start services

tests/test_engine_phase2.py

@@ -660,8 +660,7 @@ async def fake_create_pool(dsn=None):
                 store = asyncio.run(deps_mod.init_job_store())
                 assert store is mock_store
 
-    def test_falls_back_to_sqlite_when_postgres_unavailable(self):
-        from app.infra.jobs.sqlite_job_store import SQLiteJobStore
+    def test_raises_when_database_url_set_but_postgres_unavailable(self):
         import app.adapters.http.deps as deps_mod
 
         async def fail(*a, **kw):
@@ -672,5 +671,5 @@ async def fail(*a, **kw):
                 "app.infra.jobs.postgres_job_store.PostgresJobStore.create_pool",
                 side_effect=fail,
             ):
-                store = asyncio.run(deps_mod.init_job_store())
-                assert isinstance(store, SQLiteJobStore)
+                with pytest.raises(RuntimeError, match="DATABASE_URL is set but Postgres is unreachable"):
+                    asyncio.run(deps_mod.init_job_store())