fix: 性能测试 — 修复进度条卡在0%、中止任务误报崩溃、增加 Tokenizer Path 字段

evalscope/service/blueprints/eval.py

@@ -13,6 +13,7 @@
     DEFAULT_MULTIMODAL_BENCHMARKS,
     DEFAULT_TEXT_BENCHMARKS,
     OUTPUT_DIR,
+    TaskStoppedError,
     build_benchmark_entry,
     create_log_file,
     discover_all_benchmarks,
@@ -163,6 +164,9 @@ def _execute_task(task_id: str, task_config: TaskConfig, label: str = 'Task'):
             'result': serialize_result(result),
             'table': table_str
         })
+    except TaskStoppedError:
+        logger.info(f'[{task_id}] {label} stopped by user.')
+        return jsonify({'status': 'stopped', 'task_id': task_id})
     except Exception as e:
         logger.error(f'[{task_id}] {label} failed: {e}')
         return jsonify({'status': 'error', 'task_id': task_id, 'error': str(e)}), 500

evalscope/service/blueprints/perf.py

@@ -11,6 +11,7 @@
 from ..perf_archive import PerfArchiveError
 from ..utils import (
     OUTPUT_DIR,
+    TaskStoppedError,
     create_log_file,
     get_log_content,
     run_in_subprocess,
@@ -104,6 +105,9 @@ def run_performance_test():
             'result': serialize_result(result),
             'table': table_str
         })
+    except TaskStoppedError:
+        logger.info(f'[{task_id}] Task stopped by user.')
+        return jsonify({'status': 'stopped', 'task_id': task_id})
     except Exception as e:
         logger.error(f'[{task_id}] Task failed: {e}')
         return jsonify({'status': 'error', 'task_id': task_id, 'error': str(e)}), 500

evalscope/service/utils/__init__.py

@@ -9,7 +9,14 @@
     parse_benchmark_description,
 )
 from .log import OUTPUT_DIR, create_log_file, get_log_content, validate_task_id
-from .process import run_eval_wrapper, run_in_subprocess, run_perf_wrapper, serialize_result, stop_process
+from .process import (
+    TaskStoppedError,
+    run_eval_wrapper,
+    run_in_subprocess,
+    run_perf_wrapper,
+    serialize_result,
+    stop_process,
+)
 
 __all__ = [
     'OUTPUT_DIR',
@@ -21,6 +28,7 @@
     'serialize_result',
     'stop_process',
     'run_in_subprocess',
+    'TaskStoppedError',
     'DEFAULT_TEXT_BENCHMARKS',
     'DEFAULT_MULTIMODAL_BENCHMARKS',
     'build_benchmark_entry',

evalscope/service/utils/process.py

@@ -23,6 +23,26 @@
 
 _active_lock = threading.Lock()
 
+# ---------------------------------------------------------------------------
+# Track tasks stopped by the user so we can distinguish intentional stop
+# from a genuine subprocess crash.
+# ---------------------------------------------------------------------------
+
+_user_stopped_tasks: set[str] = set()
+"""Task IDs that were explicitly stopped by the user via /stop."""
+
+
+class TaskStoppedError(Exception):
+    """Raised when a task is explicitly stopped by the user.
+
+    This is a *normal* termination, not a crash — callers should handle it
+    gracefully (e.g. return ``{'status': 'stopped'}`` instead of 500).
+    """
+
+    def __init__(self, task_id: str):
+        self.task_id = task_id
+        super().__init__(f'Task {task_id} was stopped by the user.')
+
 
 def register_process(task_id: str, proc: multiprocessing.Process) -> None:
     """Register a running subprocess so it can be stopped later."""
@@ -45,12 +65,21 @@ def stop_process(task_id: str) -> bool:
         proc = _active_processes.pop(task_id, None)
     if proc is None:
         return False
+    if not proc.is_alive():
+        # Process already exited on its own; do NOT mark as user-stopped, or a
+        # pending success result would be masked as TaskStoppedError. Treat the
+        # stop request as a no-op.
+        logger.info(f'Task {task_id} already finished before stop; ignoring stop request.')
+        return False
+    # Mark as user-stopped BEFORE terminating: the parent thread blocked in
+    # run_in_subprocess polls the child every 0.1s and would otherwise observe
+    # the death first and misreport it as a crash.
+    _user_stopped_tasks.add(task_id)
+    proc.terminate()
+    proc.join(timeout=3)
     if proc.is_alive():
-        proc.terminate()
-        proc.join(timeout=3)
-        if proc.is_alive():
-            proc.kill()
-            proc.join(timeout=2)
+        proc.kill()
+        proc.join(timeout=2)
     logger.info(f'Task {task_id} stopped by user.')
     return True
 
@@ -137,27 +166,40 @@ def run_in_subprocess(func, *args, task_id=None, **kwargs):
     if task_id:
         unregister_process(task_id)
 
+    # Consume the user-stop marker exactly once so it never leaks, regardless
+    # of which branch below handles the outcome.
+    user_stopped = bool(task_id) and task_id in _user_stopped_tasks
+    if task_id:
+        _user_stopped_tasks.discard(task_id)
+
+    if res is None:
+        # The child exited without the poll loop seeing a result. Do one final
+        # non-blocking drain in case the item arrived between the last loop
+        # iteration and p.join() returning — a produced result always takes
+        # priority over the user-stop marker.
+        try:
+            res = result_queue.get_nowait()
+        except queue.Empty:
+            res = None
+
     if res is not None:
         if res['status'] == 'error':
+            # A user-initiated stop may surface as an error result when the
+            # child converts SIGTERM into an exception — still a graceful stop.
+            if user_stopped:
+                raise TaskStoppedError(task_id)
             stderr_info = res.get('stderr', '')
             stderr_section = f'\n[stderr]\n{stderr_info}' if stderr_info.strip() else ''
             raise RuntimeError(f"Subprocess error: {res['error']}\n{res.get('traceback', '')}{stderr_section}")
         return res['result']
 
+    # If the task was explicitly stopped by the user, treat it as a graceful
+    # termination — not a crash.
+    if user_stopped:
+        raise TaskStoppedError(task_id)
+
     # res is still None: the child exited without putting anything in the queue
     # (OOM, SIGKILL, import error, segfault, etc.).
-    # Do one final non-blocking check in case the item arrived between the last
-    # loop iteration and p.join() returning.
-    try:
-        res = result_queue.get_nowait()
-        if res['status'] == 'error':
-            stderr_info = res.get('stderr', '')
-            stderr_section = f'\n[stderr]\n{stderr_info}' if stderr_info.strip() else ''
-            raise RuntimeError(f"Subprocess error: {res['error']}\n{res.get('traceback', '')}{stderr_section}")
-        return res['result']
-    except queue.Empty:
-        pass
-
     raise RuntimeError(
         f'Subprocess terminated unexpectedly (exit code {p.exitcode}). '
         'The child process may have crashed due to OOM, a missing import, '

evalscope/web/src/api/schemas/eval.schema.ts

@@ -29,6 +29,8 @@ export const progressResponseSchema = z
   .object({
     percent: z.number(),
     current_step: z.string().optional(),
+    /** Tracker lifecycle state written by ProgressTracker: running | completed | error. */
+    status: z.string().optional(),
   })
   .catchall(z.unknown())
 

evalscope/web/src/components/perf/PerfConfigForm.tsx

@@ -1,9 +1,11 @@
 import { useState, type SyntheticEvent } from 'react'
 import { useLocale } from '@/contexts/LocaleContext'
 import Button from '@/components/ui/Button'
+import Card from '@/components/ui/Card'
 import Field from '@/components/ui/Field'
 import { FORM_INPUT_CLASS, inputClass } from '@/components/ui/formStyles'
 import { computeFirstInvalid, validateNumeric, FORM_MESSAGE_KEYS } from '@/domain/form/validation'
+import { ChevronDown, ChevronUp } from 'lucide-react'
 import { useFormErrors } from '@/hooks/useFormErrors'
 import { ApiKeyField, ApiUrlField, ModelField } from '@/components/tasks/TaskFormFields'
 
@@ -26,6 +28,7 @@ const IDS = {
   dataset: 'perf-dataset',
   maxPromptLen: 'perf-maxPromptLen',
   minPromptLen: 'perf-minPromptLen',
+  tokenizerPath: 'perf-tokenizerPath',
 } as const
 
 /** DOM order of focusable fields, drives first-invalid focus on submit. */
@@ -42,6 +45,7 @@ const DOM_ORDER: string[] = [
   IDS.dataset,
   IDS.maxPromptLen,
   IDS.minPromptLen,
+  IDS.tokenizerPath,
 ]
 
 export default function PerfConfigForm({ onSubmit, disabled }: Props) {
@@ -58,6 +62,8 @@ export default function PerfConfigForm({ onSubmit, disabled }: Props) {
   const [dataset, setDataset] = useState('')
   const [maxPromptLen, setMaxPromptLen] = useState('')
   const [minPromptLen, setMinPromptLen] = useState('')
+  const [tokenizerPath, setTokenizerPath] = useState('')
+  const [showMore, setShowMore] = useState(false)
 
   const { setErrors, errorFor: errMsg, clearError: clearErr } = useFormErrors()
 
@@ -119,6 +125,7 @@ export default function PerfConfigForm({ onSubmit, disabled }: Props) {
     if (dataset) config.dataset = dataset
     if (maxPromptLen) config.max_prompt_length = Number(maxPromptLen)
     if (minPromptLen) config.min_prompt_length = Number(minPromptLen)
+    if (tokenizerPath) config.tokenizer_path = tokenizerPath
     onSubmit(config)
   }
 
@@ -194,6 +201,28 @@ export default function PerfConfigForm({ onSubmit, disabled }: Props) {
         </Field>
       </div>
 
+      {/* ── More params toggle ── */}
+      <button
+        type="button"
+        onClick={() => setShowMore(!showMore)}
+        className="flex items-center gap-1 text-xs text-[var(--accent)] hover:underline cursor-pointer"
+      >
+        {t('perf.moreParams')}
+        {showMore ? <ChevronUp size={14} /> : <ChevronDown size={14} />}
+      </button>
+
+      {showMore && (
+        <Card className="!p-0">
+          <div className="grid grid-cols-1 md:grid-cols-2 gap-4 p-4">
+            <Field id={IDS.tokenizerPath} name="tokenizer_path" labelKey="perf.tokenizerPath">
+              {(aria) => (
+                <input {...aria} type="text" value={tokenizerPath} onChange={(e) => setTokenizerPath(e.target.value)} className={FORM_INPUT_CLASS} placeholder="/path/to/tokenizer" />
+              )}
+            </Field>
+          </div>
+        </Card>
+      )}
+
       <Button type="submit" variant="primary" disabled={disabled} className="btn-glow">
         {t('perf.startPerf')}
       </Button>

evalscope/web/src/components/tasks/TaskRunnerPage.tsx

@@ -91,7 +91,11 @@ export default function TaskRunnerPage({
     interval: 5000,
     onData: (data) => {
       setProgress(data.percent ?? 0)
-      if (data.percent >= 100) setRunning(false)
+      // Stop polling when benchmark finishes (via percent) or when the
+      // backend progress tracker reports error / completed (status key).
+      if (data.percent >= 100 || data.status === 'error' || data.status === 'completed') {
+        setRunning(false)
+      }
     },
   })
 

evalscope/web/src/i18n/translations/perf.ts

@@ -12,6 +12,8 @@ export const en: Dict = {
   dataset: 'Test Dataset',
   maxPromptLen: 'Max Prompt Length',
   minPromptLen: 'Min Prompt Length',
+  moreParams: 'More Parameters',
+  tokenizerPath: 'Tokenizer Path',
   startPerf: 'Start Performance Test',
   status: 'Status & Logs',
   ready: 'Status: Ready',
@@ -29,6 +31,8 @@ export const zh: Dict = {
   dataset: '测试数据集',
   maxPromptLen: '最大Prompt长度',
   minPromptLen: '最小Prompt长度',
+  moreParams: '更多参数',
+  tokenizerPath: 'Tokenizer 路径',
   startPerf: '开始性能测试',
   status: '运行状态与日志',
   ready: '当前状态: 准备就绪',

tests/service/test_process_stop.py

@@ -0,0 +1,84 @@
+# Copyright (c) Alibaba, Inc. and its affiliates.
+"""Tests for user-initiated stop handling in service subprocess utilities.
+
+Covers the interaction between ``run_in_subprocess`` and ``stop_process``:
+a user-initiated stop must surface as :class:`TaskStoppedError` (graceful),
+while a genuine child crash must still raise ``RuntimeError``.
+"""
+import multiprocessing
+import threading
+import time
+import unittest
+
+from evalscope.service.utils.process import (
+    TaskStoppedError,
+    _user_stopped_tasks,
+    register_process,
+    run_in_subprocess,
+    stop_process,
+)
+
+
+def _sleep_forever() -> str:
+    time.sleep(60)
+    return 'done'
+
+
+def _quick_success() -> str:
+    return 'ok'
+
+
+def _crash() -> None:
+    raise ValueError('boom')
+
+
+class TestProcessStop(unittest.TestCase):
+
+    def test_user_stop_raises_task_stopped_error(self):
+        """Stopping a running task raises TaskStoppedError, not RuntimeError."""
+        task_id = 'test-stop-graceful'
+        outcome = {}
+
+        def run() -> None:
+            try:
+                outcome['result'] = run_in_subprocess(_sleep_forever, task_id=task_id)
+            except Exception as e:  # noqa: BLE001 - capture for assertion
+                outcome['exc'] = e
+
+        t = threading.Thread(target=run)
+        t.start()
+        time.sleep(3)  # let the spawned child start sleeping
+
+        self.assertTrue(stop_process(task_id))
+        t.join(timeout=30)
+
+        self.assertIsInstance(outcome.get('exc'), TaskStoppedError)
+
+    def test_success_returns_result(self):
+        """A normal run returns the child's result unchanged."""
+        self.assertEqual(run_in_subprocess(_quick_success, task_id='test-stop-success'), 'ok')
+
+    def test_child_crash_still_raises_runtime_error(self):
+        """A genuine child error is not misreported as a user stop."""
+        with self.assertRaises(RuntimeError):
+            run_in_subprocess(_crash, task_id='test-stop-crash')
+
+    def test_stop_unknown_task_returns_false(self):
+        """Stopping a non-existent task is a no-op returning False."""
+        self.assertFalse(stop_process('test-stop-nonexistent'))
+
+    def test_stop_after_natural_exit_is_noop(self):
+        """Stopping an already-exited (but still registered) process must not
+        set the user-stop marker, so a pending success result is not masked."""
+        task_id = 'test-stop-already-dead'
+        proc = multiprocessing.get_context('spawn').Process(target=_quick_success)
+        proc.start()
+        proc.join()  # let it exit naturally before the stop request
+        register_process(task_id, proc)
+
+        self.assertFalse(stop_process(task_id))
+        self.assertNotIn(task_id, _user_stopped_tasks)
+
+
+if __name__ == '__main__':
+    unittest.main()