From 33c118c80e70cec64c9369b7ba4088c61c44bd31 Mon Sep 17 00:00:00 2001
From: mfyCn-1204 <1332490378@qq.com>
Date: Thu, 25 Sep 2025 14:15:02 +0800
Subject: [PATCH] [core]vllm-ascend support msMonitor tool (#3123)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

### What this PR does / why we need it?
vllm-ascend support [msMonitor
](https://gitcode.com/Ascend/mstt/tree/master/msmonitor)tool to collect
performance of vllm-ascend

### Does this PR introduce _any_ user-facing change?
1.add env MSMONITOR_USE_DAEMON；
2.user cann enable msMonitor tool by setting MSMONITOR_USE_DAEMON=1
before run vllm-ascend model；
3.MSMONITOR_USE_DAEMON and VLLM_TORCH_PROFILER_DIR cannot both set

### How was this patch tested?
1.run vllm-ascend model while not set MSMONITOR_USE_DAEMON=1 or set
MSMONITOR_USE_DAEMON=0, model will run successfully;
2.run vllm-ascend model while set MSMONITOR_USE_DAEMON=1, run msMonitor
tool to collect profile data;
3.run vllm-ascend model while set MSMONITOR_USE_DAEMON=1 and
VLLM_TORCH_PROFILER_DIR, will raise error

- vLLM version: v0.10.2
- vLLM main:
https://github.com/vllm-project/vllm/commit/f225ea7dd98e9f29752e5c032cd4a8ee1d712f16

Signed-off-by: mei-feiyao <1332490378@qq.com>
---
 tests/ut/worker/test_worker_v1.py | 22 ++++++++++++++++++++++
 vllm_ascend/envs.py               |  3 +++
 vllm_ascend/worker/worker_v1.py   | 10 ++++++++++
 3 files changed, 35 insertions(+)

diff --git a/tests/ut/worker/test_worker_v1.py b/tests/ut/worker/test_worker_v1.py
index 7ae9aa3..eb05a7a 100644
--- a/tests/ut/worker/test_worker_v1.py
+++ b/tests/ut/worker/test_worker_v1.py
@@ -355,6 +355,28 @@ class TestNPUWorker(TestBase):
 
             self.assertIn("Profiler is not enabled", str(cm.exception))
 
+    @patch("vllm_ascend.worker.worker_v1.envs_vllm")
+    @patch("vllm_ascend.worker.worker_v1.envs_ascend")
+    def test_profile_and_msmonitor_both_enabled_raises_error(
+            self, mock_envs_vllm, mock_envs_ascend):
+        """Test profile method raises exception when both profiler and msmonitor are enabled"""
+        from vllm_ascend.worker.worker_v1 import NPUWorker
+
+        mock_envs_vllm.VLLM_TORCH_PROFILER_DIR = "/path/to/traces"
+        mock_envs_ascend.MSMONITOR_USE_DAEMON = 1
+
+        # Create worker mock
+        with patch.object(NPUWorker, "__init__", lambda x, **kwargs: None):
+            worker = NPUWorker()
+
+            # Test should raise exception
+            with self.assertRaises(RuntimeError) as cm:
+                _ = worker._init_profiler()
+
+            self.assertIn(
+                "MSMONITOR_USE_DAEMON and VLLM_TORCH_PROFILER_DIR cannot be both set at the same time.",
+                str(cm.exception))
+
     def test_lora_methods(self):
         """Test LoRA related methods"""
         from vllm_ascend.worker.worker_v1 import NPUWorker
diff --git a/vllm_ascend/envs.py b/vllm_ascend/envs.py
index 61df5e1..dec0a12 100644
--- a/vllm_ascend/envs.py
+++ b/vllm_ascend/envs.py
@@ -159,6 +159,9 @@ env_variables: Dict[str, Callable[[], Any]] = {
     # caused by the initialization of the Mooncake connector.
     "PHYSICAL_DEVICES":
     lambda: os.getenv("PHYSICAL_DEVICES", None),
+    # Whether to enable msMonitor tool to monitor the performance of vllm-ascend.
+    "MSMONITOR_USE_DAEMON":
+    lambda: bool(int(os.getenv("MSMONITOR_USE_DAEMON", '0'))),
 }
 
 # end-env-vars-definition
diff --git a/vllm_ascend/worker/worker_v1.py b/vllm_ascend/worker/worker_v1.py
index 820ec63..fedec87 100644
--- a/vllm_ascend/worker/worker_v1.py
+++ b/vllm_ascend/worker/worker_v1.py
@@ -25,6 +25,7 @@ import torch.nn as nn
 import torch_npu
 import vllm.envs as envs_vllm
 from torch_npu.op_plugin.atb._atb_ops import _register_atb_extensions
+from torch_npu.profiler import dynamic_profile as dp
 from vllm.config import VllmConfig
 from vllm.distributed import (ensure_model_parallel_initialized,
                               init_distributed_environment)
@@ -41,6 +42,7 @@ from vllm.v1.outputs import (EMPTY_MODEL_RUNNER_OUTPUT, AsyncModelRunnerOutput,
                              DraftTokenIds, ModelRunnerOutput)
 from vllm.v1.worker.worker_base import WorkerBase
 
+import vllm_ascend.envs as envs_ascend
 from vllm_ascend.ascend_config import init_ascend_config
 from vllm_ascend.device_allocator.camem import CaMemAllocator
 from vllm_ascend.distributed.parallel_state import init_ascend_model_parallel
@@ -221,6 +223,10 @@ class NPUWorker(WorkerBase):
         self,
         scheduler_output: "SchedulerOutput",
     ) -> Optional[Union[ModelRunnerOutput, AsyncModelRunnerOutput]]:
+        # enable msMonitor to monitor the performance of vllm-ascend
+        if envs_ascend.MSMONITOR_USE_DAEMON:
+            dp.step()
+
         intermediate_tensors = None
         forward_pass = scheduler_output.total_num_scheduled_tokens > 0
         if forward_pass and not get_pp_group().is_first_rank:
@@ -350,6 +356,10 @@ class NPUWorker(WorkerBase):
         # Torch profiler. Enabled and configured through env vars:
         # VLLM_TORCH_PROFILER_DIR=/path/to/save/trace
         if envs_vllm.VLLM_TORCH_PROFILER_DIR:
+            if envs_ascend.MSMONITOR_USE_DAEMON:
+                raise RuntimeError(
+                    "MSMONITOR_USE_DAEMON and VLLM_TORCH_PROFILER_DIR cannot be both set at the same time."
+                )
             torch_profiler_trace_dir = envs_vllm.VLLM_TORCH_PROFILER_DIR
             logger.info("Profiling enabled. Traces will be saved to: %s",
                         torch_profiler_trace_dir)