[Bugfix] Keep VLLM_RAY_BUNDLE_INDICES out of the compile cache key (#61211)

Signed-off-by: DanBlanaru <48605845+DanBlanaru@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

tests/test_envs.py

@@ -48,6 +48,14 @@ def test_api_key_is_not_compile_factor(monkeypatch: pytest.MonkeyPatch):
     assert "VLLM_API_KEY" not in envs.compile_factors()
 
 
+def test_ray_bundle_indices_is_not_compile_factor(monkeypatch: pytest.MonkeyPatch):
+    # Engines sharing a placement group get different bundle indices but
+    # compile the same graphs, so they must share a cache key.
+    monkeypatch.setenv("VLLM_RAY_BUNDLE_INDICES", "4,5,6,7")
+
+    assert "VLLM_RAY_BUNDLE_INDICES" not in envs.compile_factors()
+
+
 def test_p2p_side_channel_defaults_and_override(monkeypatch: pytest.MonkeyPatch):
     monkeypatch.delenv("VLLM_P2P_SIDE_CHANNEL_HOST", raising=False)
     monkeypatch.delenv("VLLM_P2P_SIDE_CHANNEL_PORT", raising=False)

vllm/envs.py

@@ -2366,6 +2366,7 @@ def compile_factors() -> dict[str, object]:
         "VLLM_RANDOMIZE_DP_DUMMY_INPUTS",
         "VLLM_MODEL_REDIRECT_PATH",
         "VLLM_HOST_IP",
+        "VLLM_RAY_BUNDLE_INDICES",
         "VLLM_ELASTIC_EP_SCALE_UP_LAUNCH",
         "VLLM_FORCE_AOT_LOAD",
         "S3_ACCESS_KEY_ID",