LCORE-455: Removed global vars to make compatible with uvicorn workers > 1

src/app/endpoints/authorized.py

@@ -5,13 +5,12 @@
 
 from fastapi import APIRouter, Depends
 
-from authentication.interface import AuthTuple
 from authentication import get_auth_dependency
-from models.responses import AuthorizedResponse, UnauthorizedResponse, ForbiddenResponse
+from authentication.interface import AuthTuple
+from models.responses import AuthorizedResponse, ForbiddenResponse, UnauthorizedResponse
 
 logger = logging.getLogger(__name__)
 router = APIRouter(tags=["authorized"])
-auth_dependency = get_auth_dependency()
 
 
 authorized_responses: dict[int | str, dict[str, Any]] = {
@@ -38,7 +37,7 @@
 
 @router.post("/authorized", responses=authorized_responses)
 async def authorized_endpoint_handler(
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
 ) -> AuthorizedResponse:
     """
     Handle request to the /authorized endpoint.

src/app/endpoints/config.py

@@ -3,10 +3,10 @@
 import logging
 from typing import Annotated, Any
 
-from fastapi import APIRouter, Request, Depends
+from fastapi import APIRouter, Depends, Request
 
-from authentication.interface import AuthTuple
 from authentication import get_auth_dependency
+from authentication.interface import AuthTuple
 from authorization.middleware import authorize
 from configuration import configuration
 from models.config import Action, Configuration
@@ -15,8 +15,6 @@
 logger = logging.getLogger(__name__)
 router = APIRouter(tags=["config"])
 
-auth_dependency = get_auth_dependency()
-
 
 get_config_responses: dict[int | str, dict[str, Any]] = {
     200: {
@@ -63,7 +61,7 @@
 @router.get("/config", responses=get_config_responses)
 @authorize(Action.GET_CONFIG)
 async def config_endpoint_handler(
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
     request: Request,
 ) -> Configuration:
     """

src/app/endpoints/conversations.py

@@ -3,22 +3,21 @@
 import logging
 from typing import Any
 
+from fastapi import APIRouter, Depends, HTTPException, Request, status
 from llama_stack_client import APIConnectionError, NotFoundError
 
-from fastapi import APIRouter, HTTPException, Request, status, Depends
-
-from client import AsyncLlamaStackClientHolder
-from configuration import configuration
 from app.database import get_session
 from authentication import get_auth_dependency
 from authorization.middleware import authorize
+from client import AsyncLlamaStackClientHolder
+from configuration import configuration
 from models.config import Action
 from models.database.conversations import UserConversation
 from models.responses import (
-    ConversationResponse,
     ConversationDeleteResponse,
-    ConversationsListResponse,
     ConversationDetails,
+    ConversationResponse,
+    ConversationsListResponse,
     UnauthorizedResponse,
 )
 from utils.endpoints import (
@@ -30,7 +29,6 @@
 
 logger = logging.getLogger("app.endpoints.handlers")
 router = APIRouter(tags=["conversations"])
-auth_dependency = get_auth_dependency()
 
 conversation_responses: dict[int | str, dict[str, Any]] = {
     200: {
@@ -180,7 +178,7 @@ def simplify_session_data(session_data: dict) -> list[dict[str, Any]]:
 @authorize(Action.LIST_CONVERSATIONS)
 async def get_conversations_list_endpoint_handler(
     request: Request,
-    auth: Any = Depends(auth_dependency),
+    auth: Any = Depends(get_auth_dependency()),
 ) -> ConversationsListResponse:
     """Handle request to retrieve all conversations for the authenticated user."""
     check_configuration_loaded(configuration)
@@ -242,7 +240,7 @@ async def get_conversations_list_endpoint_handler(
 async def get_conversation_endpoint_handler(
     request: Request,
     conversation_id: str,
-    auth: Any = Depends(auth_dependency),
+    auth: Any = Depends(get_auth_dependency()),
 ) -> ConversationResponse:
     """
     Handle request to retrieve a conversation by ID.
@@ -370,7 +368,7 @@ async def get_conversation_endpoint_handler(
 async def delete_conversation_endpoint_handler(
     request: Request,
     conversation_id: str,
-    auth: Any = Depends(auth_dependency),
+    auth: Any = Depends(get_auth_dependency()),
 ) -> ConversationDeleteResponse:
     """
     Handle request to delete a conversation by ID.

src/app/endpoints/conversations_v2.py

@@ -3,25 +3,24 @@
 import logging
 from typing import Any
 
-from fastapi import APIRouter, Request, Depends, HTTPException, status
+from fastapi import APIRouter, Depends, HTTPException, Request, status
 
-from configuration import configuration
 from authentication import get_auth_dependency
 from authorization.middleware import authorize
+from configuration import configuration
 from models.cache_entry import CacheEntry
 from models.config import Action
 from models.responses import (
-    ConversationsListResponseV2,
-    ConversationResponse,
     ConversationDeleteResponse,
+    ConversationResponse,
+    ConversationsListResponseV2,
     UnauthorizedResponse,
 )
 from utils.endpoints import check_configuration_loaded
 from utils.suid import check_suid
 
 logger = logging.getLogger("app.endpoints.handlers")
 router = APIRouter(tags=["conversations_v2"])
-auth_dependency = get_auth_dependency()
 
 
 conversation_responses: dict[int | str, dict[str, Any]] = {
@@ -93,7 +92,7 @@
 @authorize(Action.LIST_CONVERSATIONS)
 async def get_conversations_list_endpoint_handler(
     request: Request,  # pylint: disable=unused-argument
-    auth: Any = Depends(auth_dependency),
+    auth: Any = Depends(get_auth_dependency()),
 ) -> ConversationsListResponseV2:
     """Handle request to retrieve all conversations for the authenticated user."""
     check_configuration_loaded(configuration)
@@ -123,7 +122,7 @@ async def get_conversations_list_endpoint_handler(
 async def get_conversation_endpoint_handler(
     request: Request,  # pylint: disable=unused-argument
     conversation_id: str,
-    auth: Any = Depends(auth_dependency),
+    auth: Any = Depends(get_auth_dependency()),
 ) -> ConversationResponse:
     """Handle request to retrieve a conversation by ID."""
     check_configuration_loaded(configuration)
@@ -159,7 +158,7 @@ async def get_conversation_endpoint_handler(
 async def delete_conversation_endpoint_handler(
     request: Request,  # pylint: disable=unused-argument
     conversation_id: str,
-    auth: Any = Depends(auth_dependency),
+    auth: Any = Depends(get_auth_dependency()),
 ) -> ConversationDeleteResponse:
     """Handle request to delete a conversation by ID."""
     check_configuration_loaded(configuration)

src/app/endpoints/feedback.py

@@ -1,12 +1,13 @@
 """Handler for REST API endpoint for user feedback."""
 
+import json
 import logging
 import threading
-from typing import Annotated, Any
+from datetime import UTC, datetime
 from pathlib import Path
-import json
-from datetime import datetime, UTC
-from fastapi import APIRouter, HTTPException, Depends, Request, status
+from typing import Annotated, Any
+
+from fastapi import APIRouter, Depends, HTTPException, Request, status
 
 from authentication import get_auth_dependency
 from authentication.interface import AuthTuple
@@ -18,15 +19,14 @@
     ErrorResponse,
     FeedbackResponse,
     FeedbackStatusUpdateResponse,
+    ForbiddenResponse,
     StatusResponse,
     UnauthorizedResponse,
-    ForbiddenResponse,
 )
 from utils.suid import get_suid
 
 logger = logging.getLogger(__name__)
 router = APIRouter(prefix="/feedback", tags=["feedback"])
-auth_dependency = get_auth_dependency()
 feedback_status_lock = threading.Lock()
 
 # Response for the feedback endpoint
@@ -87,7 +87,7 @@ async def assert_feedback_enabled(_request: Request) -> None:
 @authorize(Action.FEEDBACK)
 async def feedback_endpoint_handler(
     feedback_request: FeedbackRequest,
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
     _ensure_feedback_enabled: Any = Depends(assert_feedback_enabled),
 ) -> FeedbackResponse:
     """Handle feedback requests.
@@ -183,7 +183,7 @@ def feedback_status() -> StatusResponse:
 @authorize(Action.ADMIN)
 async def update_feedback_status(
     feedback_update_request: FeedbackStatusUpdateRequest,
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
 ) -> FeedbackStatusUpdateResponse:
     """
     Handle feedback status update requests.

src/app/endpoints/health.py

@@ -8,25 +8,23 @@
 import logging
 from typing import Annotated, Any
 
+from fastapi import APIRouter, Depends, Response, status
 from llama_stack.providers.datatypes import HealthStatus
 
-from fastapi import APIRouter, status, Response, Depends
-from client import AsyncLlamaStackClientHolder
-from authentication.interface import AuthTuple
 from authentication import get_auth_dependency
+from authentication.interface import AuthTuple
 from authorization.middleware import authorize
+from client import AsyncLlamaStackClientHolder
 from models.config import Action
 from models.responses import (
     LivenessResponse,
-    ReadinessResponse,
     ProviderHealthStatus,
+    ReadinessResponse,
 )
 
 logger = logging.getLogger("app.endpoints.handlers")
 router = APIRouter(tags=["health"])
 
-auth_dependency = get_auth_dependency()
-
 
 async def get_providers_health_statuses() -> list[ProviderHealthStatus]:
     """
@@ -80,7 +78,7 @@ async def get_providers_health_statuses() -> list[ProviderHealthStatus]:
 @router.get("/readiness", responses=get_readiness_responses)
 @authorize(Action.INFO)
 async def readiness_probe_get_method(
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
     response: Response,
 ) -> ReadinessResponse:
     """
@@ -126,7 +124,7 @@ async def readiness_probe_get_method(
 @router.get("/liveness", responses=get_liveness_responses)
 @authorize(Action.INFO)
 async def liveness_probe_get_method(
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
 ) -> LivenessResponse:
     """
     Return the liveness status of the service.

src/app/endpoints/info.py

@@ -3,24 +3,21 @@
 import logging
 from typing import Annotated, Any
 
-from fastapi import APIRouter, HTTPException, Request, status
-from fastapi import Depends
+from fastapi import APIRouter, Depends, HTTPException, Request, status
 from llama_stack_client import APIConnectionError
 
-from authentication.interface import AuthTuple
 from authentication import get_auth_dependency
+from authentication.interface import AuthTuple
 from authorization.middleware import authorize
-from configuration import configuration
 from client import AsyncLlamaStackClientHolder
+from configuration import configuration
 from models.config import Action
 from models.responses import InfoResponse
 from version import __version__
 
 logger = logging.getLogger("app.endpoints.handlers")
 router = APIRouter(tags=["info"])
 
-auth_dependency = get_auth_dependency()
-
 
 get_info_responses: dict[int | str, dict[str, Any]] = {
     200: {
@@ -40,7 +37,7 @@
 @router.get("/info", responses=get_info_responses)
 @authorize(Action.INFO)
 async def info_endpoint_handler(
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
     request: Request,
 ) -> InfoResponse:
     """

src/app/endpoints/metrics.py

@@ -1,28 +1,27 @@
 """Handler for REST API call to provide metrics."""
 
 from typing import Annotated
+
+from fastapi import APIRouter, Depends, Request
 from fastapi.responses import PlainTextResponse
-from fastapi import APIRouter, Request, Depends
 from prometheus_client import (
-    generate_latest,
     CONTENT_TYPE_LATEST,
+    generate_latest,
 )
 
-from authentication.interface import AuthTuple
 from authentication import get_auth_dependency
+from authentication.interface import AuthTuple
 from authorization.middleware import authorize
-from models.config import Action
 from metrics.utils import setup_model_metrics
+from models.config import Action
 
 router = APIRouter(tags=["metrics"])
 
-auth_dependency = get_auth_dependency()
-
 
 @router.get("/metrics", response_class=PlainTextResponse)
 @authorize(Action.GET_METRICS)
 async def metrics_endpoint_handler(
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
     request: Request,
 ) -> PlainTextResponse:
     """

src/app/endpoints/models.py

@@ -9,9 +9,9 @@
 
 from authentication import get_auth_dependency
 from authentication.interface import AuthTuple
+from authorization.middleware import authorize
 from client import AsyncLlamaStackClientHolder
 from configuration import configuration
-from authorization.middleware import authorize
 from models.config import Action
 from models.responses import ModelsResponse
 from utils.endpoints import check_configuration_loaded
@@ -20,9 +20,6 @@
 router = APIRouter(tags=["models"])
 
 
-auth_dependency = get_auth_dependency()
-
-
 models_responses: dict[int | str, dict[str, Any]] = {
     200: {
         "models": [
@@ -54,7 +51,7 @@
 @authorize(Action.GET_MODELS)
 async def models_endpoint_handler(
     request: Request,
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
 ) -> ModelsResponse:
     """
     Handle requests to the /models endpoint.

src/app/endpoints/query.py

@@ -57,7 +57,6 @@
 
 logger = logging.getLogger("app.endpoints.handlers")
 router = APIRouter(tags=["query"])
-auth_dependency = get_auth_dependency()
 
 query_response: dict[int | str, dict[str, Any]] = {
     200: {
@@ -174,7 +173,7 @@ def evaluate_model_hints(
 async def query_endpoint_handler(
     request: Request,
     query_request: QueryRequest,
-    auth: Annotated[AuthTuple, Depends(auth_dependency)],
+    auth: Annotated[AuthTuple, Depends(get_auth_dependency())],
     mcp_headers: dict[str, dict[str, str]] = Depends(mcp_headers_dependency),
 ) -> QueryResponse:
     """