Make it possible to configure how rerun_notebook fetches the widget

[jleibs]

What

• Partially addresses: Viewer instances don't release memory when cell output is cleared #6642
• Also fixes: The new self-contained widget is slow in google colab #6651

Overview

Even though rerun_notebook ships with the widget asset bundled in, by default it will try to load the asset
from https://app.rerun.io. This is because the way anywiget transmits the asset at the moment results in
a memory leak of the entire module for each cell execution.

If your network does not allow you to access app.rerun.io, the behavior can be changed by setting the
the RERUN_NOTEBOOK_ASSET environment variable before you import rerun_notebook. This variable must
be set prior to your import because AnyWidget stores the resource on the widget class instance
once at import time.

Inlined asset

Setting:

RERUN_NOTEBOOK_ASSET=inline

Will cause rerun_notebook to directly transmit the inlined asset to the widget over Jupyter comms.
This will be the most portable way to use the widget, but is currently known to leak memory and
has some performance issues in environments such as Google colab.

Locally served asset

Setting:

RERUN_NOTEBOOK_ASSET=serve-local

Will cause rerun_notebook to launch a thread serving the asset from the local machine during
the lifetime of the kernel. This will be the best way to use the widget in a notebook environment
when your notebook server is running locally.

Manually hosted asset

Setting:

RERUN_NOTEBOOK_ASSET=https://your-hosted-asset-url.com/widget.js

Will cause rerun_notebook to load the asset from the provided URL. This is the most flexible way to
use the widget, but requires you to host the asset yourself.

The rerun_notebook package has a minimal server that can be used to serve the asset nanually by running:

python -m rerun_notebook serve

However, any hosting platform can be used to serve the asset, as long as it is accessible to the notebook
and has appropriate CORS headers set. asset_server.py for a simple example.

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested the web demo (if applicable):
  • Using examples from latest main build: rerun.io/viewer
  • Using full set of examples from nightly build: rerun.io/viewer
• The PR title and labels are set such as to maximize their usefulness for the next release's CHANGELOG
• If applicable, add a new check to the release checklist!

• PR Build Summary
• Recent benchmark results
• Wasm size tracking

To run all checks from main, comment on the PR with @rerun-bot full-check.

[emilk]

Is it possible to print an error message pointing the user to RERUN_NOTEBOOK_ASSET if they fail to load the viewer from app.rerun.io?

[github-actions]

Deployed docs

Commit	Link
8d1213b	https://landing-1lm6a04il-rerun.vercel.app/docs