`Kernel._control_lock` created after it may be used

未关闭适合新手

#1,567 创建于 2026年9月22日 1 条评论 0 个 reaction 已指派 0 人在 GitHub 查看

还没有人认领这个 Issue。

评估

难度: 2/5
预计耗时: 1-3 小时
新手友好度: 75/100

Issue 类型: 缺陷
描述清晰度: 描述清楚
活跃度: 活跃
技术栈: python
领域: backend

调研方向

问题位于 ipykernel/kernelbase.py 第 347 行附近，_control_lock 在初始化之前被访问。查看内核类的 __init__ 方法以了解 _control_lock 的设置位置。修复方法是确保在分发任何控制消息之前创建锁。请参考链接分支中的补丁，然后通过在模拟的高延迟环境中运行内核进行验证。

由索引模型根据 Issue 内容生成。

描述

I ran into the following exception when running a Notebook inside VSCode with the python environment running on a remote machine (through the VSCode tunnel). The remote machine has a distributed file system which can cause significant latencies when starting a notebook from scratch, which can unearth race conditions that would be unlikely on a normal laptop or desktop machine. I guess our somewhat offbeat use case is the reason this did not get noticed before.

ERROR:tornado.general:Uncaught exception in ZMQStream callback
Traceback (most recent call last):
  File "/apps/gent/RHEL9/cascadelake-ib/software/IPython/9.14.0-GCCcore-15.2.0/lib/python3.14/site-packages/traitlets/traitlets.py", line 651, in get
    value = obj._trait_values[self.name]
            ~~~~~~~~~~~~~~~~~^^^^^^^^^^^
KeyError: '_control_lock'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/apps/gent/RHEL9/cascadelake-ib/software/PyZMQ/27.1.0-GCCcore-15.2.0/lib/python3.14/site-packages/zmq/eventloop/zmqstream.py", line 565, in _log_error
    f.result()
    ~~~~~~~~^^
  File "/apps/gent/RHEL9/cascadelake-ib/software/jupyter-server/2.19.0-GCCcore-15.2.0/lib/python3.14/site-packages/ipykernel/kernelbase.py", line 347, in dispatch_control
    async with self._control_lock:
               ^^^^^^^^^^^^^^^^^^
  File "/apps/gent/RHEL9/cascadelake-ib/software/IPython/9.14.0-GCCcore-15.2.0/lib/python3.14/site-packages/traitlets/traitlets.py", line 706, in __get__
    return self.get(obj, cls)  # type:ignore[return-value]
           ~~~~~~~~^^^^^^^^^^
  File "/apps/gent/RHEL9/cascadelake-ib/software/IPython/9.14.0-GCCcore-15.2.0/lib/python3.14/site-packages/traitlets/traitlets.py", line 668, in get
    value = self._validate(obj, default)
  File "/apps/gent/RHEL9/cascadelake-ib/software/IPython/9.14.0-GCCcore-15.2.0/lib/python3.14/site-packages/traitlets/traitlets.py", line 741, in _validate
    value = self.validate(obj, value)
  File "/apps/gent/RHEL9/cascadelake-ib/software/IPython/9.14.0-GCCcore-15.2.0/lib/python3.14/site-packages/traitlets/traitlets.py", line 2319, in validate
    self.error(obj, value)
    ~~~~~~~~~~^^^^^^^^^^^^
  File "/apps/gent/RHEL9/cascadelake-ib/software/IPython/9.14.0-GCCcore-15.2.0/lib/python3.14/site-packages/traitlets/traitlets.py", line 850, in error
    raise TraitError(e)
traitlets.traitlets.TraitError: The '_control_lock' trait of an IPythonKernel instance expected a Lock, not the NoneType None.

This was with ipykernel v7.2.0 on Redhat Linux 9 (remote machine, HPC infrastructure). I've written a patch for v7.2.0, so we can perform a minimal upgrade of our teaching setup:

https://github.com/tovrstra/ipykernel/tree/bugfix/control_lock_race

I think the same issue (_control_lock created too late) is still present in the main branch. I'll make a PR on the main branch in minute.