Using Stlite instead of a hosted Python runtime naturally limits the usability of Streamlit. I’m curious to what degree the following limitations have been considered and may be mitigated:Query parameters: my app use query parameters to store the state and enable easy sharing of configurations. Will Streamlit on CDF ever support this? Session state, persistence: How should things persisted when all is lost on a simple refresh? I don’t image writing state to CDF would work well. Memory: I’ve worked with larger data sets of which only summaries are shown to the user. Could cognite functions work as a backend for this? 3rd party libraries: Pydantic 2, Pola.rs &c. can’t compile to WASM (yet). I don’t image there is anything you can do about this, but I mention it because it is probably the most significant limitation and I’m curious as to whatever thoughts you have on it.

Seasoned ⭐️

Question

Regarding current stlite limitations

Forum|Forum|2 years ago
May 3, 2024
4 replies
98 views

Using Stlite instead of a hosted Python runtime naturally limits the usability of Streamlit. I’m curious to what degree the following limitations have been considered and may be mitigated:

Query parameters: my app use query parameters to store the state and enable easy sharing of configurations. Will Streamlit on CDF ever support this?
Session state, persistence: How should things persisted when all is lost on a simple refresh? I don’t image writing state to CDF would work well.
Memory: I’ve worked with larger data sets of which only summaries are shown to the user. Could cognite functions work as a backend for this?
3rd party libraries: Pydantic 2, Pola.rs &c. can’t compile to WASM (yet). I don’t image there is anything you can do about this, but I mention it because it is probably the most significant limitation and I’m curious as to whatever thoughts you have on it.

Øystein Aspøy

Seasoned ⭐️⭐️

@Armin Gharakhanloo Maybe this is something you can look into?

Lars Moastuen

Expert ⭐️⭐️⭐️⭐️

Hi @Boye Sjo and sorry for the delayed reply.

I’ll go through your questions one by one:

Query parameters

This is actually possible to do today using st.query_params:

import streamlit as st
my_param = st.query_params['myCustomParameter']
st.write(my_param)

You will need to add the parameter to the URL. It might also be an idea to ensure that parameters hideCode=true and hideToolbar=true is set to hide code environment.

Session state

For session state we recommend st.session_state. This allows you to store per-session state, but it does not serialize across sessions. If you wish to store state between sessions you could look into https://pypi.org/project/extra-streamlit-components/ to store state across sessions using browser storage. I have not tried this myself but would be interested in hearing your experience if you try it out.

Memory

I think using Cognite Functions as a way to offload heavy computation is an interesting approach and urge you to try :)

3rd party libraries

I agree that this is probably the most severe limitation of an in-browser implementation of Streamlit. We are very much aware of this and are discussing if/how to mitigate some of the pains. One thing we have discussed is the possibility to embed locally built “Python wheels” (*.whl) in the app to at least enable building wheels locally when they are not available.

I hope this helps :)

Lars Moastuen (Software Engineer, Atlas AI)

Dilini Fernando

Expert ⭐️⭐️⭐️⭐️

Hi @Boye Sjo,

Did the above help you?

B

Boye SjoAuthor

Seasoned ⭐️

Somewhat. I can not get query parameters to work. They are accessible from

st.query_params

but do not appear in the browser’s URL bar and do not persist when switching back from another page.

As it stands, for the purposes of my current projects. I don’t think the native CDF hosting and user authentication will be worth the effort of migrating away from Pola.rs, query parameter states and server-wide caching in addition the risk of worse performance.

Sign up

Welcome to Cognite Hub