Re: ipython/jupyter issue [Was: RuntimeError: Kernel died before replying to kernel_info (#4116)]
On 10/11/17 17:28, Andreas Tille wrote:
> Hi,
>
> On Thu, Nov 09, 2017 at 08:48:02PM +0100, Gordon Ball wrote:
>>
>> I just built statsmodels 0.8.0-6 in and amd64 sbuild chroot without
>> encountering this issue.
>>
>> Looking at the trace in #880245, it looks like the key error is
>>
>> zmq.error.ZMQError: Address already in use
>>
>> and the other
>>
>> RuntimeError: Kernel died before replying to kernel_info
>>
>> are just caused by the other process not being able to communicate. My
>> guess is that this is an environment heisenbug - container network
>> issues/socket exhaustion...
>
> IMHO this would be a good reason to lower the severity of the bug, isn't
> it?
You can check this for yourself, but yes - it appears that it does not
consistently FTBFS, and therefore shouldn't be RC.
>
>> It might be possible to get `ExecutePreprocessor` in
>> `tools/nbgenerate.py` to use IPC instead of TCP transport, which might
>> be more reliable for this sort of use (although it probably doesn't get
>> tested as much, so might have its own problems); this is a supported
>> option when invoking some CLI tools (eg, `jupyter console --transport
>> ipc --kernel python3`), but it's not obvious to me how to do it in this
>> case.
>
> Any further hits?
I don't think I have time to dig into this at the moment, but given the
nbgenerate task is already run (docs/Makefile:58,70) with
`--allow_errors=True` to ignore errors within the notebooks, I'd just
patch the Makefile to ignore any errors from that command, which in the
worst case will result in some documentation notebooks not being populated.
>
> Kind regards
>
> Andreas.
>
Reply to: