[CI] Add pre-commit check for patch logger (#7446)

### What this PR does / why we need it?
See https://github.com/vllm-project/vllm-ascend/pull/7402, pre-commit
hook will forbid init_logger(__name__) in vllm_ascend patch modules

- vLLM version: v0.17.0
- vLLM main:
8a680463fa

---------

Signed-off-by: wangli <wangli858794774@gmail.com>

.pre-commit-config.yaml

@@ -91,6 +91,12 @@ repos:
     language: python
     types: [python]
     pass_filenames: false
+  - id: check-logger
+    name: Forbid init_logger(__name__) in vllm_ascend modules
+    entry: tools/check_logger.sh
+    language: script
+    types: [python]
+    pass_filenames: false
   # Keep `suggestion` last
   - id: suggestion
     name: Suggestion

tests/ut/quantization/test_quant_utils.py

@@ -1,5 +1,4 @@
 import json
-import logging
 import os
 import tempfile
 from pathlib import Path
@@ -135,16 +134,16 @@ class TestMaybeAutoDetectQuantization(TestBase):
         vllm_config = self._make_vllm_config(
             model_path="/fake/quant_model", quantization=None)
 
-        with self.assertLogs("vllm_ascend.quantization.utils",
+        with patch("vllm_ascend.quantization.utils.logger") as mock_logger:
-                             level=logging.INFO) as cm:
             maybe_auto_detect_quantization(vllm_config)
 
         self.assertEqual(vllm_config.model_config.quantization,
                          ASCEND_QUANTIZATION_METHOD)
-        log_output = "\n".join(cm.output)
+        mock_logger.info.assert_called_once()
-        self.assertIn("Auto-detected quantization method", log_output)
+        call_args = mock_logger.info.call_args[0]
-        self.assertIn(ASCEND_QUANTIZATION_METHOD, log_output)
+        self.assertIn("Auto-detected quantization method", call_args[0])
-        self.assertIn("/fake/quant_model", log_output)
+        self.assertIn(ASCEND_QUANTIZATION_METHOD, call_args)
+        self.assertIn("/fake/quant_model", call_args)
 
     @patch("vllm_ascend.quantization.utils.detect_quantization_method",
            return_value=ASCEND_QUANTIZATION_METHOD)
@@ -155,29 +154,28 @@ class TestMaybeAutoDetectQuantization(TestBase):
             model_path="/fake/quant_model",
             quantization=COMPRESSED_TENSORS_METHOD)
 
-        with self.assertLogs("vllm_ascend.quantization.utils",
+        with patch("vllm_ascend.quantization.utils.logger") as mock_logger:
-                             level=logging.WARNING) as cm:
             maybe_auto_detect_quantization(vllm_config)
 
         self.assertEqual(vllm_config.model_config.quantization,
                          COMPRESSED_TENSORS_METHOD)
-        log_output = "\n".join(cm.output)
+        mock_logger.warning.assert_called_once()
-        self.assertIn("Auto-detected quantization method", log_output)
+        call_args = mock_logger.warning.call_args[0]
-        self.assertIn(ASCEND_QUANTIZATION_METHOD, log_output)
+        self.assertIn("Auto-detected quantization method", call_args[0])
-        self.assertIn(COMPRESSED_TENSORS_METHOD, log_output)
+        self.assertIn(ASCEND_QUANTIZATION_METHOD, call_args)
+        self.assertIn(COMPRESSED_TENSORS_METHOD, call_args)
 
     @patch("vllm_ascend.quantization.utils.detect_quantization_method",
            return_value=None)
     def test_no_detection_emits_no_log(self, mock_detect):
         """When no quantization is detected, no log should be emitted."""
         vllm_config = self._make_vllm_config(quantization=None)
-        logger_name = "vllm_ascend.quantization.utils"
 
-        with self.assertRaises(AssertionError):
+        with patch("vllm_ascend.quantization.utils.logger") as mock_logger:
-            # assertLogs raises AssertionError when no logs are emitted
+            maybe_auto_detect_quantization(vllm_config)
-            with self.assertLogs(logger_name, level=logging.DEBUG):
-                maybe_auto_detect_quantization(vllm_config)
 
+        mock_logger.info.assert_not_called()
+        mock_logger.warning.assert_not_called()
         self.assertIsNone(vllm_config.model_config.quantization)
 
     @patch("vllm.config.VllmConfig._get_quantization_config",

tools/check_logger.sh

@@ -0,0 +1,69 @@
+#!/bin/bash
+#
+# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# This file is a part of the vllm-ascend project.
+#
+# Check that vllm_ascend modules do not use init_logger(__name__).
+#
+# vllm's logging config registers a handler only for the "vllm" logger
+# namespace.  Any logger created via init_logger(__name__) inside a
+# vllm_ascend module ends up in the "vllm_ascend.*" namespace, which has
+# no handler, so every log call is silently dropped.
+#
+# The correct pattern is:
+#   from vllm.logger import logger
+#
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+REPO_ROOT="$(cd "$SCRIPT_DIR/.." && pwd)"
+PATCH_DIR="$REPO_ROOT/vllm_ascend/"
+
+VIOLATIONS=0
+
+for FILE in $(find "$PATCH_DIR" -type f -name "*.py" 2>/dev/null); do
+    [[ -f "$FILE" ]] || continue
+
+    # Find lines that call init_logger(__name__)
+    while IFS= read -r MATCH; do
+        LINENUM=$(echo "$MATCH" | cut -d: -f1)
+        LINE=$(echo "$MATCH" | cut -d: -f2-)
+        if [[ $VIOLATIONS -eq 0 ]]; then
+            echo ""
+        fi
+        echo "  $FILE:$LINENUM: $LINE"
+        VIOLATIONS=$(( VIOLATIONS + 1 ))
+    done < <(grep -n 'init_logger[[:space:]]*([[:space:]]*__name__[[:space:]]*)' "$FILE" 2>/dev/null || true)
+done
+
+if [[ $VIOLATIONS -gt 0 ]]; then
+    echo ""
+    echo "Found $VIOLATIONS violation(s): init_logger(__name__) must not be used in vllm_ascend modules."
+    echo ""
+    echo "vllm's logging handler is registered only for the 'vllm' namespace."
+    echo "Loggers created with init_logger(__name__) inside vllm_ascend end up"
+    echo "in the 'vllm_ascend.*' namespace, which has no handler — all log"
+    echo "messages are silently dropped."
+    echo ""
+    echo "Fix: replace"
+    echo "   from vllm.logger import init_logger"
+    echo "   logger = init_logger(__name__)"
+    echo "with"
+    echo "   from vllm.logger import logger"
+    exit 1
+fi
+
+exit 0

vllm_ascend/_310p/quantization/modelslim_config.py

@@ -21,7 +21,7 @@ from typing import Any
 
 import torch
 from vllm.config import get_current_vllm_config
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.linear import LinearBase
 from vllm.model_executor.layers.quantization import register_quantization_config
@@ -41,8 +41,6 @@ from vllm_ascend.quantization.modelslim_config import (
 )
 from vllm_ascend.utils import ASCEND_QUANTIZATION_METHOD
 
-logger = init_logger(__name__)
-
 
 def create_scheme_for_layer(
     quant_description: dict[str, Any],

vllm_ascend/batch_invariant.py

@@ -20,14 +20,13 @@ import os
 
 import torch
 import torch_npu
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.model_executor.layers.batch_invariant import vllm_is_batch_invariant
 from vllm.triton_utils import HAS_TRITON
 
 # in case recursive call in reduce_sum.
 torch_sum = torch.sum
 
-logger = init_logger(__name__)
 
 if HAS_TRITON:
     from vllm_ascend.ops.triton.batch_invariant.matmul import (

vllm_ascend/core/recompute_scheduler.py

@@ -27,7 +27,7 @@ from vllm.distributed.ec_transfer.ec_connector.base import ECConnectorMetadata
 from vllm.distributed.kv_events import KVEventBatch
 from vllm.distributed.kv_transfer.kv_connector.v1.base import KVConnectorMetadata
 from vllm.distributed.kv_transfer.kv_connector.v1.metrics import KVConnectorStats
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.v1.core.kv_cache_manager import KVCacheBlocks
 from vllm.v1.core.sched.async_scheduler import AsyncScheduler
 from vllm.v1.core.sched.output import NewRequestData, SchedulerOutput
@@ -42,8 +42,6 @@ from vllm.v1.sample.rejection_sampler import PLACEHOLDER_TOKEN_ID
 from vllm.v1.spec_decode.metrics import SpecDecodingStats
 from vllm.v1.utils import ConstantList, record_function_or_nullcontext
 
-logger = init_logger(__name__)
-
 
 # `spec_manager_map` in single_type_kv_cache_manager is a module-level dict
 # whose keys are class objects bound at import time.  When the async

vllm_ascend/distributed/kv_transfer/kv_pool/ucm_connector.py

@@ -5,11 +5,8 @@ import torch
 from ucm.integration.vllm.ucm_connector import UCMConnector
 from vllm.config import VllmConfig
 from vllm.distributed.kv_transfer.kv_connector.v1.base import KVConnectorBase_V1, KVConnectorMetadata, KVConnectorRole
-from vllm.logger import init_logger
 from vllm.v1.core.sched.output import SchedulerOutput
 
-logger = init_logger(__name__)
-
 # isort: off
 if TYPE_CHECKING:
     from vllm.v1.attention.backend import AttentionMetadata  # type: ignore

vllm_ascend/kv_offload/cpu_npu.py

@@ -1,13 +1,11 @@
 import numpy as np
 import torch
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.utils.platform_utils import is_pin_memory_available
 from vllm.v1.attention.backend import AttentionBackend  # type: ignore
 from vllm.v1.kv_offload.mediums import CPULoadStoreSpec, GPULoadStoreSpec
 from vllm.v1.kv_offload.worker.worker import OffloadingHandler, TransferResult, TransferSpec
 
-logger = init_logger(__name__)
-
 
 def expand_block_ids(
     block_ids: np.ndarray,

vllm_ascend/patch/platform/patch_mamba_config_310.py

@@ -4,7 +4,7 @@
 from math import lcm
 
 import vllm.model_executor.models.config
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.model_executor.models import ModelRegistry
 from vllm.model_executor.models.config import MambaModelConfig
 from vllm.utils.math_utils import cdiv
@@ -24,7 +24,6 @@ def verify_and_update_config(cls, vllm_config) -> None:
     Args:
         vllm_config: vLLM Config
     """
-    logger = init_logger(__name__)
     # Save the user input before it gets modified by MambaModelConfig
     mamba_block_size = vllm_config.cache_config.mamba_block_size
     # Enable FULL_AND_PIECEWISE by default

vllm_ascend/quantization/compressed_tensors_config.py

@@ -21,7 +21,7 @@ from typing import Any, Optional, cast
 
 import torch
 from compressed_tensors.quantization import QuantizationArgs, QuantizationStrategy, QuantizationType
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.linear import LinearBase, UnquantizedLinearMethod
 from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS, register_quantization_config
@@ -37,8 +37,6 @@ from vllm_ascend.utils import COMPRESSED_TENSORS_METHOD
 
 from .methods import AscendLinearScheme, AscendMoEScheme
 
-logger = init_logger(__name__)
-
 
 # Remove the original compressed_tensors method to replace with our implementation
 def _remove_quantization_method():

vllm_ascend/quantization/modelslim_config.py

@@ -31,7 +31,7 @@ from typing import Any, Optional
 
 import torch
 from vllm.config import get_current_vllm_config
-from vllm.logger import init_logger
+from vllm.logger import logger
 from vllm.model_executor.layers.attention_layer_base import AttentionLayerBase
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.linear import LinearBase
@@ -47,8 +47,6 @@ from .methods import get_scheme_class
 # The config filename that ModelSlim generates after quantizing a model.
 MODELSLIM_CONFIG_FILENAME = "quant_model_description.json"
 
-logger = init_logger(__name__)
-
 # key: model_type
 # value: vLLM prefix -> HF prefix mapping (used to convert vLLM layer names to HF format
 # for looking up keys in quant_model_description.json)

vllm_ascend/quantization/utils.py

@@ -19,12 +19,10 @@ import json
 from pathlib import Path
 
 from vllm import envs
-from vllm.logger import init_logger
+from vllm.logger import logger
 
 from vllm_ascend.utils import ASCEND_QUANTIZATION_METHOD, COMPRESSED_TENSORS_METHOD
 
-logger = init_logger(__name__)
-
 
 def get_model_file(
     model: str | Path,

vllm_ascend/spec_decode/draft_proposer.py

@@ -2,14 +2,11 @@ import torch
 import torch.nn as nn
 from typing_extensions import override
 from vllm.config import VllmConfig
-from vllm.logger import init_logger
 from vllm.model_executor.model_loader import get_model
 from vllm.v1.spec_decode.utils import create_vllm_config_for_draft_model
 
 from vllm_ascend.spec_decode.eagle_proposer import SpecDecodeBaseProposer
 
-logger = init_logger(__name__)
-
 
 class AscendDraftModelProposer(SpecDecodeBaseProposer):
     def __init__(

vllm_ascend/spec_decode/medusa_proposer.py

@@ -1,14 +1,11 @@
 import torch
 from vllm.config import CUDAGraphMode
-from vllm.logger import init_logger
 from vllm.v1.sample.metadata import SamplingMetadata
 from vllm.v1.spec_decode.medusa import MedusaProposer
 from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
 
 from vllm_ascend.ascend_forward_context import set_ascend_forward_context
 
-logger = init_logger(__name__)
-
 
 class AscendMedusaProposer(MedusaProposer):
     """