This gets us closer to logging parts of the request
as separate json fields but not all the way there.
For example:
```json
{"written_at": "2020-10-07T21:52:16.975Z", "written_ts": 1602107536975497000, "msg": "{'status': 404, 'method': 'GET', 'ip': '127.0.0.1', 'uri': '/nbextensions/widgets/notebook/js/extension.js', 'request_time': 1.9807815551757812, 'referer': 'http://localhost:8888/notebooks/test.ipynb'}", "type": "log", "logger": "NotebookApp", "thread": "MainThread", "level": "WARNING", "module": "log", "line_no": 54}
```
This adds a new extra package dependency on `json-logging`
and an environment variable, which when set to "true" regardless
of case, will try to use the json-logging non-web app log formatter.
If the `json-logging` package is not installed but the environment
variable is True, something like this will be logged but it will not
crash the application:
```
$ ENABLE_JSON_LOGGING=true jupyter notebook
Unable to use json logging due to missing packages. Run "pip install notebook[json-logging]" to fix.
Traceback (most recent call last):
File "/home/osboxes/jupyter/notebook/notebook/notebookapp.py", line 144, in <module>
import json_logging
ModuleNotFoundError: No module named 'json_logging'
```
Initially I tried to add a new Bool config option to toggle this but
the problem is (from my limited understanding of traitlets and tornado)
is that `_log_formatter_cls` needs to be set early and trying to se the
log formatter later in `init_logging` is too late - or at least I couldn't
figure out a way to reset the log formatter dynamically (I tried calling
`_log_format_changed` [1] but that didn't work).
With this working you get logs like this:
```
{"written_at": "2020-10-07T21:10:51.606Z", "written_ts": 1602105051606265000, "msg": "404 GET /nbextensions/widgets/notebook/js/extension.js (127.0.0.1) 9.26ms referer=http://localhost:8888/notebooks/Untitled.ipynb", "type": "log", "logger": "NotebookApp", "thread": "MainThread", "level": "WARNING", "module": "log", "line_no": 49}
{"written_at": "2020-10-07T21:10:54.443Z", "written_ts": 1602105054443309000, "msg": "Starting buffering for f260ddd7-938c-42d0-ac3b-455bea76694f:49f30b53fc4b4ec6a8f2fb748a171613", "type": "log", "logger": "NotebookApp", "thread": "MainThread", "level": "INFO", "module": "kernelmanager", "line_no": 222}
{"written_at": "2020-10-07T21:10:54.446Z", "written_ts": 1602105054446264000, "msg": "Kernel shutdown: f260ddd7-938c-42d0-ac3b-455bea76694f", "type": "log", "logger": "NotebookApp", "thread": "MainThread", "level": "INFO", "module": "multikernelmanager", "line_no": 201}
```
An obvious improvement here would to be able to split the `msg` fields apart
so that we can log things like response status code, request method, request
URL, response_time_ms etc. That should be possible with the `JSONLogWebFormatter`
from `json-logging` but when I tried using that I was getting errors from
the library about a correlation id (which is based on a request header we
don't use). The `json-logging` library supports several web frameworks like
Flask but unfortunately does not have built in support for Tornado, but it does
support custom formatters so that might be a follow up option to improve on this.
[1] https://github.com/ipython/traitlets/blob/4.3.3/traitlets/config/application.py#L195Closes#5798
Since the guide isn't using the environment.yml file anymore
it doesn't make a lot of sense to link to a doc about building
a conda environment from a yml file.
With the `docs` install extra in setup.py we can simplify the
setup for building the docs by avoiding the now redundant
docs/doc-requirements.txt and docs/environment.yml.
One change is needed in setup.py though which is to add the
`sphinx-rtd-theme` package to the setup.py `docs` extra install
package list which was in doc-requirements.txt and environment.yml
but missing from setup.py.
To build the docs you need the notebook package installed from
the local project so this updates the contributor guide for the
conda instructions.
Related #5741
The instructions on building the docs fail to mention that
the notebook package itself needs to be installed. I was following
the pip-based instructions so this fixes that case. I'm not familiar
with using conda really so that case isn't fixed here.
Partial #5741
The section `Notebook front-end client` in the config overview
page was essentially a toc tree except the sub-section links
went to the top of the frontend_config page rather than the real
sub-sections in that page because they all used the same reference.
Rather than create a unique reference for each sub-section in the
frontend_config page, this simply embeds the frontend_config toctree
within the config_overview page which essentially has the same effect.
Closes#5740
- Change the scope of styles specific to `.dynamic-buttons` to that class only (introed in https://github.com/jupyter/notebook/pull/4729)
- Drop the width style from tooltips since `display: inline` makes it have no effect
Fixes#5670