[Bugfix] Break duplex OpenAI import cycle (#8287)

Signed-off-by: Nick Cao <ncao@redhat.com>
Co-authored-by: Codex <noreply@openai.com>

tests/entrypoints/openai_api/test_api_server_guards.py

@@ -58,8 +58,7 @@ from starlette.requests import Request
 from starlette.websockets import WebSocketDisconnect
 from vllm.v1.engine.exceptions import EngineDeadError, EngineGenerateError
 
-from vllm_omni.entrypoints.duplex import openai as duplex_openai
-from vllm_omni.entrypoints.openai import api_server
+from vllm_omni.entrypoints.openai import api_server, realtime_connection
 from vllm_omni.entrypoints.serve.utils import errors as serve_errors
 
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
@@ -675,7 +674,11 @@ async def test_realtime_route_defaults_to_configured_duplex_handler(
         async def handle_connection(self) -> None:
             calls.append("legacy")
 
-    monkeypatch.setattr(duplex_openai, "RealtimeConnection", lambda _websocket, _serving: _LegacyConnection())
+    monkeypatch.setattr(
+        realtime_connection,
+        "RealtimeConnection",
+        lambda _websocket, _serving: _LegacyConnection(),
+    )
     query_params = {} if duplex_query is None else {"duplex": duplex_query}
     websocket = SimpleNamespace(
         app=SimpleNamespace(

vllm_omni/entrypoints/duplex/openai.py

@@ -12,8 +12,6 @@ from vllm.logger import init_logger
 from vllm.utils import random_uuid
 
 from vllm_omni.entrypoints.duplex.warmup import DUPLEX_WARMUP_CLIENT_WAIT_S
-from vllm_omni.entrypoints.openai.realtime.connection import OpenAIFullDuplexConnection
-from vllm_omni.entrypoints.openai.realtime_connection import RealtimeConnection
 
 if TYPE_CHECKING:
     from vllm_omni.config.omni_config import BaseVllmOmniStageConfig
@@ -23,6 +21,9 @@ logger = init_logger(__name__)
 _QWEN3_OMNI_REALTIME_ARCH = "Qwen3OmniMoeForConditionalGeneration"
 _QWEN3_OMNI_REALTIME_STAGES = {"thinker", "talker", "code2wav"}
 
+# Keep OpenAI connection imports in the handlers: openai/__init__.py imports
+# api_server, which imports this module.
+
 
 def supports_qwen3_omni_realtime(stage_configs: Sequence[BaseVllmOmniStageConfig] | None) -> bool:
     stages = stage_configs or ()
@@ -59,6 +60,8 @@ async def dispatch_generic_realtime_websocket(websocket: WebSocket) -> None:
         await websocket.send_json({"type": "error", "error": "Realtime API is not available", "code": "unsupported"})
         await websocket.close()
         return
+    from vllm_omni.entrypoints.openai.realtime_connection import RealtimeConnection
+
     connection = RealtimeConnection(websocket, serving)
     await connection.handle_connection()
 
@@ -93,6 +96,8 @@ async def dispatch_realtime_websocket(websocket: WebSocket) -> None:
             )
 
     tokenizer = await state.engine_client.get_tokenizer()
+    from vllm_omni.entrypoints.openai.realtime.connection import OpenAIFullDuplexConnection
+
     connection = OpenAIFullDuplexConnection(
         websocket=websocket,
         engine=state.engine_client,