[Fix] Add warning for incompatible Ray backend with ACL Graph mode (#1501)

### What this PR does / why we need it?

Currently, Ray is not compatible with ACL Graph, so we need to fall back
to eager mode when using the Ray backend.

### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

---------

Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>

Author: yiz-liu

tests/singlecard/test_aclgraph.py

@@ -103,3 +103,16 @@ def test_deepseek_raises_error(monkeypatch: pytest.MonkeyPatch) -> None:
                        max_model_len=1024,
                        enforce_eager=False)
         assert "ACL Graph does not support deepseek" in str(excinfo.value)
+
+
+@pytest.mark.skipif(os.getenv("VLLM_USE_V1") == "0",
+                    reason="aclgraph only support on v1")
+@pytest.mark.parametrize("model", MODELS)
+def test_ray_backend_sets_no_compilation(
+        model: str, monkeypatch: pytest.MonkeyPatch) -> None:
+    with monkeypatch.context() as m:
+        m.setenv("VLLM_USE_V1", "1")
+        runner = VllmRunner(model,
+                            enforce_eager=False,
+                            distributed_executor_backend="ray")
+        assert runner.model.llm_engine.vllm_config.compilation_config.level == 0

vllm_ascend/platform.py

@@ -157,6 +157,11 @@ def check_and_update_config(cls, vllm_config: VllmConfig) -> None:
                 "Torchair compilation enabled on NPU. Setting level to NO_COMPILATION"
             )
             compilation_config.level = CompilationLevel.NO_COMPILATION
+        elif parallel_config.distributed_executor_backend == "ray":
+            logger.warning(
+                "Ray distributed executor backend is not compatible with ACL Graph mode "
+                "right now. Setting level to NO_COMPILATION")
+            compilation_config.level = CompilationLevel.NO_COMPILATION
         else:
             logger.info(
                 "PIECEWISE compilation enabled on NPU. use_inductor not supported - "