fix(monitor): 抖音任务永远「运行中」—— page.evaluate 卡在一个死掉的标签页上，而我没给超时

用户报「一直在运行中」。库里那条 run 是真的卡住了：日志里一条采集输出都没有，说明它卡在 collect 里、还没走到任何日志。探针定位到：标签页: [''] ← 一个 URL 为空的标签页，渲染进程已卡死 context.cookies(): OK, 90 个 ← cookie 读得到 page.evaluate('navigator.userAgent'): **永远不返回** 而问浏览器要身份（UA + client hints）是采集的**第一步**，`page.evaluate` 又**没设超时** —— 于是整轮挂在那儿，run 永远停在「运行中」。三处修复，各挡一层： 1. `page.evaluate` / `context.cookies()` 全部加超时（8 秒）。卡住就跳过，不再无限等。 2. 不假设第一个标签页是好的：逐个试、优先抖音页；全都不行就临时开一个干净页问完关掉。拿不到就退回库里那份 cookie —— **不编造指纹**，那比没有更糟。 3. **进程内那条路补上整体超时**：爬虫那条靠 `run_and_wait(timeout=...)` 兜底，这条路没有子进程、没人管，里面任何一次卡住都会变成永久的「运行中」。测试 +6：卡死的页会被跳过（真 sleep，验的正是超时）、没 UA 的页跳过、全不行时开临时页并关掉它、优先抖音页；以及整轮卡住时 run 不会停在 running（含超时原因）。
2026-10-10 17:51:54 +08:00
parent 9e13a7f686
commit 5552e2a2b8
5 changed files with 1087 additions and 17 deletions
@@ -116,6 +116,101 @@ class TestNormalizeAweme:
        assert record["create_time"] == 0


+class TestIdentityFromPages:
+    """身份得从浏览器里问，但**不能被一个卡死的标签页拖住**。
+
+    实测过：标签页 URL 为空、渲染进程卡死，``page.evaluate`` 永远不返回；而问身份是采集的
+    第一步 —— 没超时的话整轮就挂在那儿，run 永远停在「运行中」。
+    """
+
+    class _Page:
+        def __init__(self, url, *, user_agent=None, hang=False):
+            self.url = url
+            self._user_agent = user_agent
+            self._hang = hang
+            self.closed = False
+
+        async def evaluate(self, expression):
+            if self._hang:
+                await asyncio.sleep(30)  # 模拟渲染进程卡死
+            if expression.startswith("() => navigator.userAgentData"):
+                return {
+                    "brands": [{"brand": "Chrome", "version": "155"}],
+                    "mobile": False,
+                    "platform": "Linux",
+                }
+            return self._user_agent
+
+        async def close(self):
+            self.closed = True
+
+    class _Context:
+        def __init__(self, pages, temp=None):
+            self.pages = pages
+            self._temp = temp
+            self.made_temp = False
+
+        async def new_page(self):
+            self.made_temp = True
+            if self._temp is None:
+                raise AssertionError("这个用例不该走到临时页")
+            return self._temp
+
+    def _fast_timeout(self, monkeypatch):
+        monkeypatch.setattr(douyin_api, "EVALUATE_TIMEOUT_SECONDS", 0.05)
+
+    @pytest.mark.asyncio
+    async def test_a_hanging_page_is_skipped(self, monkeypatch):
+        self._fast_timeout(monkeypatch)
+        stuck = self._Page("", hang=True)
+        good = self._Page("https://example.com/", user_agent="UA-of-good-page")
+
+        user_agent, hints = await douyin_api._identity_from_pages(
+            self._Context([stuck, good])
+        )
+
+        assert user_agent == "UA-of-good-page"
+        assert hints["sec-ch-ua"] == '"Chrome";v="155"'
+
+    @pytest.mark.asyncio
+    async def test_a_page_without_a_user_agent_is_skipped(self, monkeypatch):
+        self._fast_timeout(monkeypatch)
+        blank = self._Page("", user_agent=None)
+        good = self._Page("https://example.com/", user_agent="UA-of-good-page")
+
+        user_agent, _ = await douyin_api._identity_from_pages(
+            self._Context([blank, good])
+        )
+
+        assert user_agent == "UA-of-good-page"
+
+    @pytest.mark.asyncio
+    async def test_it_opens_a_temporary_page_when_nothing_else_works(self, monkeypatch):
+        self._fast_timeout(monkeypatch)
+        stuck = self._Page("", hang=True)
+        temp = self._Page("about:blank", user_agent="UA-of-temp-page")
+        context = self._Context([stuck], temp=temp)
+
+        user_agent, _ = await douyin_api._identity_from_pages(context)
+
+        assert context.made_temp is True
+        assert user_agent == "UA-of-temp-page"
+        assert temp.closed is True, "临时页问完要关掉，别在操作者的浏览器里留垃圾"
+
+    @pytest.mark.asyncio
+    async def test_a_douyin_page_is_preferred(self, monkeypatch):
+        """有抖音页面就先问它 —— 它才是我们要模仿的那个身份。"""
+        self._fast_timeout(monkeypatch)
+        other = self._Page("https://example.com/", user_agent="UA-of-other")
+        douyin = self._Page("https://www.douyin.com/explore", user_agent="UA-of-douyin")
+
+        user_agent, _ = await douyin_api._identity_from_pages(
+            self._Context([other, douyin])
+        )
+
+        assert user_agent == "UA-of-douyin"
+
+
 class TestGet:
    """`_get` 的失败路径 —— 它们决定了失败会不会被伪装成「这个博主没作品」。"""

@@ -5,6 +5,8 @@
 约定它一个都不沾。凡是写在那里面的东西，这条路都得单独有一份。
 """

+import asyncio
+
 import pytest
 import pytest_asyncio
 from sqlalchemy import select
@@ -80,3 +82,29 @@ class TestDouyinRunStatus:
        await runner_module.execute_task(task_id, trigger="manual")

        assert seen["status"] == RUN_RUNNING
+
+    @pytest.mark.asyncio
+    async def test_a_hanging_collect_does_not_leave_the_run_running(self, db, monkeypatch):
+        """进程内那条路也要有超时。
+
+        爬虫那条靠 ``run_and_wait(timeout=...)`` 兜底，这条路没有子进程、没人管 ——
+        里面任何一次卡住（实测过 ``page.evaluate`` 打在一个卡死的标签页上不返回）都会让
+        run 永远停在「运行中」，界面上看起来就是任务卡死了。
+        """
+        task_id = await _make_douyin_task()
+        async with monitor_db.get_session() as session:
+            task = await session.get(MonitorTask, task_id)
+            task.run_timeout_seconds = 1  # 把超时压到 1 秒，别让测试真等
+
+        async def hanging_collect(out_dir, **kwargs):
+            await asyncio.sleep(60)
+            raise AssertionError("不该走到这里")
+
+        monkeypatch.setattr(runner_module.douyin_fetch, "collect", hanging_collect)
+
+        await runner_module.execute_task(task_id, trigger="manual")
+
+        async with monitor_db.get_session() as session:
+            run = await session.scalar(select(MonitorRun).order_by(MonitorRun.id))
+            assert run.status != RUN_RUNNING
+            assert "超时" in (run.error_message or "") or "超过" in (run.error_message or "")