jupyter/jupyter_client

8.x timeout bug

mlucool opened this issue · 2 comments

mlucool commented

When upgrading to 7.4.9 to 8.1.0, the timeout parameter no longer works with papermill + a custom KernelManager.

Example notebook, example_slow.ipynb:

from time import sleep
sleep(10)

Executing with papermill:

import papermill as pm
pm.execute_notebook("example_slow.ipynb", "example_slow.out.ipynb", timeout=1, kernel_manager_class="jupyter_client.manager.KernelManager")

With 7.4.9 this correctly raises CellTimeoutError. With 8.1.0 the notebook runs to completion (in 10 seconds). Oddly, if I don't use a custom manager class, it also errors correctly. I need a custom manager given #154 (comment).

Is this a bug in this package or should how timeout is used be changed in papermill?

tuncbkose commented

It seems to me that jupyter_client>=8 breaks some assumptions that nbclient made, so this is not necessarily a bug with jupyter_client.

Edit: How about using jupyter_client.manager.AsyncKernelManager?

mlucool commented

This seems to work, and solves my use case. I'll close this, but I'm not sure if others feel there is still a bug here. If so, please reopen.