force set VLLM_USE_V1

Signed-off-by: Linkun Chen <github@lkchen.net>

Author: lk-chen

python/ray/llm/_internal/serve/deployments/llm/vllm/vllm_engine.py

@@ -315,6 +315,11 @@ async def _start_engine(self) -> "EngineClient":
         # 2. If VLLM_USE_V1 is set to 1, then it will use v1 engine even with
         #    experimental features (such as launching vLLM on a non-main thread).
         # 3. If VLLM_USE_V1 is set to 0, force using v0 engine.
+        if not envs.is_set("VLLM_USE_V1"):
+            raise AssertionError(
+                "Starting ray 2.45, VLLM_USE_V1 environmetn variable must be set "
+                "to prevent undetermined behavior"
+            )
         if not envs.VLLM_USE_V1:
             return await self._start_engine_v0()
         return await self._start_engine_v1()