Configuration: notebook idle timeouts, and upload limits

[Norman Gray]

Greetings.

I'm having difficulty pulling various bits of documentation together.

I need to increase the maximum size of files uploaded through the
JupyterHub interface, and I need to add an idle-timeout to sessions. I
can't work out with confidence how to do that.

Regarding timeouts: the docs at
<https://jupyterhub.readthedocs.io/en/stable/reference/config-user-env.html?highlight=timeout>
suggest adjusting the configuration in
/etc/jupyter/jupyter_notebook_config.py. But (a) I don't have that
file, (b) I can't find an authoritative statement of what files
notebooks search for their configuration (I can't find any 'jupyter' or
'jupyterhub' manpages), and (c) the documentation indicates that this
will only work for notebooks after 5.4.

I see from `/usr/bin/jupyter --version` that I'm actually using v4.3.0.
I'm doing this on the most recent version of CentOS (which does tend to
be... conservative about versions). I'm sort of reluctant to use a
version other than the OS-packaged version -- I've been through Python
package-hell before. It's do-able, of course, but not much fun.

The page <https://jupyter-core.readthedocs.io/en/latest/> seems to refer
to jupyter_core 4.5 as the most recent version. I see from
<https://jupyter.readthedocs.io/en/latest/architecture/visual_overview.html>
(a _very_ useful page!) that this is the version that relates to
/usr/bin/jupyter (so perhaps I _am_ reasonably up to date). But how do
I find what version of 'notebook' I'm running (I'm assuming that 5.4
refers to the notebook version rather than the jupyter_core version)?
I'm really confused at this point.

Supposing that my version 4.3.0 _is_ too early for the configuration
mentioned in the above config page, is there really no way of
configuring notebooks to time out? Failing that, is there a reliable
way of determining idle servers from outside, and a recommended way of
killing them in that case? (parsing `ps` output and SIGINTing the idle
ones is... *ick*... but looking suddenly attractive).

I have possibly the same question with respect to file uploads. Looking
at <https://github.com/jupyter/notebook/issues/96> there does seem to
have been an issue with the upload of 'big' files, where 35MB appears to
be regarded as 'big' (!). The history of this specific issue suggests
that this is resolved in notebooks from v5.0. Am I just out of luck
with older versions?

Thanks for any pointers.

Best wishes,

Norman


--
Norman Gray : https://nxg.me.uk

[Norman Gray]

Greetings

On 3 Oct 2018, at 14:16, Norman Gray wrote:

> Supposing that my version 4.3.0 _is_ too early for the configuration
> mentioned in the above config page, is there really no way of
> configuring notebooks to time out? Failing that, is there a reliable
> way of determining idle servers from outside, and a recommended way of
> killing them in that case? (parsing `ps` output and SIGINTing the
> idle ones is... *ick*... but looking suddenly attractive).

Focusing on just this question, is there a way of shutting down a
JupyterHub notebook 'nicely' from outside, so that any unsaved work is
saved. My goals are to be able to:

* reap 'abandoned' sessions in a way which doesn't risk losing work
(what often happens is that a user will close a tab and walk away,
without realising that this just leaves the notebook running idly); and

* similarly shut down a JupyterHub server, immediately prior to a
reboot.

Looking at the logs in the latter situation, I see a large number of
lines like

... received signal 15, stopping
... received signal 15, stopping
... Shutting down 5 kernels
... received signal 15, stopping
... Shutting down 3 kernels
... Shutting down 2 kernels
... Shutting down 5 kernels

suggesting that SIGTERM at least is being caught and _something_
well-behaved is happening. But it's not clear to me what.

Is there a Best Practice here?

Best wishes

Norman