[CI][BugFix] Fix failure CI "amd-v1-sample-plus-logits-mi300-1"

tests/v1/logits_processors/test_custom_offline.py

@@ -1,5 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from pathlib import Path
 from typing import Any
 
 import pytest
@@ -15,9 +16,9 @@
     CustomLogitprocSource,
     DummyLogitsProcessor,
     WrappedPerReqLogitsProcessor,
+    install_dummy_logitproc_entrypoint,
     prompts,
 )
-from tests.v1.logits_processors.utils import entry_points as fake_entry_points
 from vllm import LLM, SamplingParams
 from vllm.v1.sample.logits_processor import (
     STR_POOLING_REJECTS_LOGITSPROCS,
@@ -102,7 +103,11 @@ def _run_test(kwargs: dict, logitproc_loaded: bool) -> None:
 
 @create_new_process_for_each_test()
 @pytest.mark.parametrize("logitproc_source", list(CustomLogitprocSource))
-def test_custom_logitsprocs(monkeypatch, logitproc_source: CustomLogitprocSource):
+def test_custom_logitsprocs(
+    monkeypatch: pytest.MonkeyPatch,
+    tmp_path: Path,
+    logitproc_source: CustomLogitprocSource,
+) -> None:
     """Test offline Python interface for passing custom logitsprocs
 
     Construct an `LLM` instance which loads a custom logitproc that has a
@@ -145,13 +150,7 @@ def test_custom_logitsprocs(monkeypatch, logitproc_source: CustomLogitprocSource
 
     if logitproc_source == CustomLogitprocSource.LOGITPROC_SOURCE_ENTRYPOINT:
         # Scenario: vLLM loads a logitproc from a preconfigured entrypoint
-        # To that end, mock a dummy logitproc entrypoint
-        import importlib.metadata
-
-        importlib.metadata.entry_points = fake_entry_points  # type: ignore
-
-        # fork is required for workers to see entrypoint patch
-        monkeypatch.setenv("VLLM_WORKER_MULTIPROC_METHOD", "fork")
+        install_dummy_logitproc_entrypoint(monkeypatch, tmp_path)
         _run_test({}, logitproc_loaded=True)
         return
 
@@ -167,7 +166,7 @@ def test_custom_logitsprocs(monkeypatch, logitproc_source: CustomLogitprocSource
 
 
 @create_new_process_for_each_test()
-def test_custom_logitsprocs_req(monkeypatch):
+def test_custom_logitsprocs_req(monkeypatch: pytest.MonkeyPatch) -> None:
     """Test passing request-level logits processor to offline Python interface
 
     Wrap a request-level logits processor to create a batch level logits
@@ -210,8 +209,11 @@ def test_custom_logitsprocs_req(monkeypatch):
     ],
 )
 def test_rejects_custom_logitsprocs(
-    monkeypatch, model_scenario: str, logitproc_source: CustomLogitprocSource
-):
+    monkeypatch: pytest.MonkeyPatch,
+    tmp_path: Path,
+    model_scenario: str,
+    logitproc_source: CustomLogitprocSource,
+) -> None:
     """Validate that vLLM engine initialization properly rejects custom
     logitsprocs when the model is a pooling model or speculative decoding
     enabled.
@@ -266,14 +268,7 @@ def test_rejects_custom_logitsprocs(
         # Scenario: vLLM loads a model and ignores a logitproc that is
         # available at a preconfigured entrypoint
 
-        # Patch in dummy logitproc entrypoint
-        import importlib.metadata
-
-        importlib.metadata.entry_points = fake_entry_points  # type: ignore
-
-        # fork is required for entrypoint patch to be visible to workers,
-        # although they should ignore the entrypoint patch anyway
-        monkeypatch.setenv("VLLM_WORKER_MULTIPROC_METHOD", "fork")
+        install_dummy_logitproc_entrypoint(monkeypatch, tmp_path)
 
         llm = LLM(**llm_kwargs)
         # Require that no custom logitsprocs have been loaded

tests/v1/logits_processors/test_custom_online.py

@@ -4,6 +4,8 @@
 import os
 import random
 import sys
+from collections.abc import AsyncIterator, Iterator
+from pathlib import Path
 from typing import Any
 
 import openai
@@ -17,9 +19,9 @@
     MAX_TOKENS,
     MODEL_NAME,
     TEMP_GREEDY,
+    install_dummy_logitproc_entrypoint,
     prompts,
 )
-from tests.v1.logits_processors.utils import entry_points as fake_entry_points
 
 
 def _server_with_logitproc_entrypoint(
@@ -28,15 +30,8 @@ def _server_with_logitproc_entrypoint(
     vllm_serve_args: list[str],
 ) -> None:
     """Start vLLM server, inject dummy logitproc entrypoint"""
-
-    # Patch `entry_points` to inject logitproc entrypoint
-    import importlib.metadata
-
-    importlib.metadata.entry_points = fake_entry_points  # type: ignore
     from vllm.entrypoints.cli import main
 
-    # fork is required for workers to see entrypoint patch
-    os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "fork"
     if env_dict is not None:
         os.environ.update(env_dict)
 
@@ -62,7 +57,7 @@ def _server_with_logitproc_fqcn(
 
 
 @pytest.fixture(scope="module")
-def default_server_args():
+def default_server_args() -> list[str]:
     return [
         # use half precision for speed and memory savings in CI environment
         "--dtype",
@@ -77,7 +72,12 @@ def default_server_args():
 @pytest.fixture(
     scope="function", params=[[], ["--logits-processors", DUMMY_LOGITPROC_FQCN]]
 )
-def server(default_server_args, request, monkeypatch):
+def server(
+    default_server_args: list[str],
+    request: pytest.FixtureRequest,
+    monkeypatch: pytest.MonkeyPatch,
+    tmp_path: Path,
+) -> Iterator[RemoteOpenAIServerCustom]:
     """Consider two server configurations:
     (1) --logits-processors cli arg specifies dummy logits processor via fully-
     qualified class name (FQCN); patch in a dummy logits processor module
@@ -93,7 +93,8 @@ def server(default_server_args, request, monkeypatch):
         args = default_server_args + request.param
         _server_fxn = _server_with_logitproc_fqcn
     else:
-        # Launch server, inject dummy logitproc entrypoint
+        # Launch server with a discoverable dummy logitproc entrypoint
+        install_dummy_logitproc_entrypoint(monkeypatch, tmp_path)
         args = default_server_args
         _server_fxn = _server_with_logitproc_entrypoint
 
@@ -102,7 +103,9 @@ def server(default_server_args, request, monkeypatch):
 
 
 @pytest_asyncio.fixture
-async def client(server):
+async def client(
+    server: RemoteOpenAIServerCustom,
+) -> AsyncIterator[openai.AsyncOpenAI]:
     async with server.get_async_client() as async_client:
         yield async_client
 
@@ -124,7 +127,7 @@ async def client(server):
     "model_name",
     [MODEL_NAME],
 )
-def test_custom_logitsprocs(server, model_name: str):
+def test_custom_logitsprocs(server: RemoteOpenAIServerCustom, model_name: str) -> None:
     """Test custom logitsprocs when starting OpenAI server from CLI
 
     Launch vLLM OpenAI-compatible server, configured to load a custom logitproc

tests/v1/logits_processors/utils.py

@@ -1,10 +1,13 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
+import os
 import types
 from enum import Enum, auto
+from pathlib import Path
 from typing import Any
 
+import pytest
 import torch
 
 from vllm.config import VllmConfig
@@ -189,3 +192,27 @@ def new_req_logits_processor(
 
 """Fake version of importlib.metadata.entry_points"""
 entry_points = lambda group: EntryPoints(group)
+
+
+def install_dummy_logitproc_entrypoint(
+    monkeypatch: pytest.MonkeyPatch, tmp_path: Path
+) -> None:
+    """Install a temporary entrypoint visible to spawned worker processes."""
+    dist_info = tmp_path / "dummy_vllm_logitproc-0.0.dist-info"
+    dist_info.mkdir()
+    (dist_info / "METADATA").write_text(
+        "Metadata-Version: 2.1\nName: dummy-vllm-logitproc\nVersion: 0.0\n",
+        encoding="utf-8",
+    )
+    (dist_info / "entry_points.txt").write_text(
+        f"[{LOGITSPROCS_GROUP}]\n"
+        f"{DUMMY_LOGITPROC_ENTRYPOINT} = {DUMMY_LOGITPROC_FQCN}\n",
+        encoding="utf-8",
+    )
+
+    monkeypatch.syspath_prepend(str(tmp_path))
+    pythonpath = os.environ.get("PYTHONPATH")
+    monkeypatch.setenv(
+        "PYTHONPATH",
+        os.pathsep.join(filter(None, [str(tmp_path), pythonpath])),
+    )