JupyterLite

[NicolasCARPi]

https://jupyterlite.readthedocs.io/en/latest/

Basically it's jupyter notbooks that runs entirely in the browser. Could be interesting for elab. Uploaded python files or ipynb could be executed directly in the browser.

[pierrepo]

Good idea but one should be explicit on which packages (and version) are provided.

[HenningTimm]

This would definitely be of interest to several groups I spoke with.

Do I understand correctly that if this runs in the browser it is executed completely on the client machine? Then we don't need to worry about a notebook gobbling up computational resources on our eLab server, which would be very nice :)

[NicolasCARPi]

Exactly, we live in a world where everything is javascript now. Even python :p

[NicolasCARPi]

There is also a pretty nice security aspect to not having to run untrusted code on a server!

[pierrepo]

Indeed, JupyterLite soly runs in the browser. The remote server only provided a bunch of html, css, js and wasm files as any regular web server.

[anargam]

There is also a pretty nice security aspect to not having to run untrusted code on a server!

That's true!

I have not looked at this in any detail, but how would the introduction of such a feature (possibly optional, based on what the sysadmin allows for a certain team?) affect the security properties in general? If it is to be useful, how isolated can it be from eg the state of the eLabFTW experiment containing it? For instance, dynamically retrieving or saving data to an experiment sounds like it would require us to throw out a lot of the guarantees that come from disallowing allowing inline scripts, loading them etc.

[MrWhiteABEX]

There is also a pretty nice security aspect to not having to run untrusted code on a server!

Well I have to say, good for the server but potentially very bad for the user. I think that should be explicitly enabled by a user and you need a secure whitelist of packages and allowed url if external communication is possible.

[vfreysz]

Maybe one way to do it with very little changes is to :
-run jupyterlite with elabapi_python in his own container.
-unable the REPL under the editor part of experiment in elabftw experiment with :

<iframe   src="https://jupyterlite.github.io/demo/repl/index.html"   width="100%"   height="100%"
></iframe>

See https://jupyterlite.readthedocs.io/en/latest/quickstart/embed-repl.html

-Automatically save the current ipython notebook in the experiment with the api :
uploadsApi.post_upload('experiments', EXP_ID, file="the_current_notebook.ipynb")
The saved notebook could be automatically reloaded when editing.
Some function could be added to automate import csv files from elabftw experiment in jupyterlite, or figure exportation in the current experiment using the api.
with something like:

FILE = "DATA.CSV"
for upload in uploadsApi.read_uploads('experiments', exp.id):
    print(upload.id, upload.real_name, upload.comment)
    if (upload.real_name == FILE):
        FILE_ID = upload.id
open(FILE, 'wb') as csvfile:
     csvfile.write(uploadsApi.read_upload('experiments', EXP_ID, FILE_ID, format='binary', _preload_content=False).data)

fig.write_image("graph.png") 
uploadsApi.post_upload('experiments', EXP_ID, file="graph.png) 
os.remove("graph.png")

I'll be happy to help adding this great feature.

[vfreysz]

Hello,
I was trying to get APIv2 working with jupyterlite but i have an issue with multiprocessing calling by elabapi-python. Is it possible to have a flag to avoid multiprocessing calling when using configuration ?

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
Cell In[9], line 1
----> 1 charge_api(cle_de_api,API_HOST_URL,91)

Cell In[8], line 11, in charge_api(API_KEY, API_HOST_URL, EXP_ID)
      8 configuration.verify_ssl = False
     10 # create an instance of the API class
---> 11 api_client = elabapi_python.ApiClient(configuration)
     12 # fix issue with Authorization header not being properly set by the generated lib
     13 api_client.set_default_header(header_name='Authorization', header_value=API_KEY)

File /lib/python3.11/site-packages/elabapi_python/api_client.py:68, in ApiClient.__init__(self, configuration, header_name, header_value, cookie)
     65     configuration = Configuration()
     66 self.configuration = configuration
---> 68 self.pool = ThreadPool()
     69 self.rest_client = rest.RESTClientObject(configuration)
     70 self.default_headers = {}

File /lib/python311.zip/multiprocessing/pool.py:930, in ThreadPool.__init__(self, processes, initializer, initargs)
    929 def __init__(self, processes=None, initializer=None, initargs=()):
--> 930     Pool.__init__(self, processes, initializer, initargs)

File /lib/python311.zip/multiprocessing/pool.py:196, in Pool.__init__(self, processes, initializer, initargs, maxtasksperchild, context)
    192 self._taskqueue = queue.SimpleQueue()
    193 # The _change_notifier queue exist to wake up self._handle_workers()
    194 # when the cache (self._cache) is empty or when there is a change in
    195 # the _state variable of the thread that runs _handle_workers.
--> 196 self._change_notifier = self._ctx.SimpleQueue()
    197 self._cache = _PoolCache(notifier=self._change_notifier)
    198 self._maxtasksperchild = maxtasksperchild

File /lib/python311.zip/multiprocessing/context.py:112, in BaseContext.SimpleQueue(self)
    110 def SimpleQueue(self):
    111     '''Returns a queue object'''
--> 112     from .queues import SimpleQueue
    113     return SimpleQueue(ctx=self.get_context())

File /lib/python311.zip/multiprocessing/queues.py:23
     19 import errno
     21 from queue import Empty, Full
---> 23 import _multiprocessing
     25 from . import connection
     26 from . import context

ModuleNotFoundError: The module '_multiprocessing' is removed from the Python standard library in the Pyodide distribution due to browser limitations.
See https://pyodide.org/en/stable/usage/loading-packages.html for more details.```

[NicolasCARPi]

The version is pinned because latest version simply doesn't work: swagger-api/swagger-codegen#12321

So once they fix the image, it can be updated.

[vfreysz]

Good news, the commit has been pushed ;)

[NicolasCARPi]

@vfreysz Can I let you track the release and open an issue on the elabapi-python repo once it can be updated please? I don't really have the bandwidth to track this myself.

[NicolasCARPi]

@vfreysz Nevermind, it's already part of the latest release. I have updated elabapi-python to 5.0.2, it contains the fix.

[vfreysz]

Unfortunately, multithreading is still initialized. I tried to look at the swagger-codegen code and I have difficulty understanding where the code which generates the initialization is located, to propose a starting point for a patch...

for info, https://pypi.org/project/jupyterlab-spreadsheet-editor/ is now working with jupyterlite. So we can now easily enter tables and plot them directly in the browser.