Hello folks! A few months ago, we released Resoto Notebook, a library that makes it easy to query, visualize, and analyze Resoto data using pandas, Plotly, and Jupyter Notebooks.
Today, we'll discuss Resoto's new JupyterLite support, which allows you to use Notebooks in the browser without installing or launching a Jupyter server.
Want to analyze raw infrastructure data when only platform engineers can access cloud consoles? Or count infrastructure assets without a data scientist? JupyterLite is a JupyterLab distribution that runs entirely in a web browser, and Resoto's JupyterLite support gives you access to popular data analysis tools without the need for any additional installation steps.
You can access JupyterLite simply by opening https://<Resoto Core hostname or IP address>/notebook in your web browser!
The browser-based JupyterLab distribution is a bit different from the desktop version: it is not possible to use pip directly. Instead, there is a package called piplite which can install Python wheels in a similar fashion.
Since JupyterLite works (almost) the same as a regular Jupyter Notebook, there are many cool things we can do with it. Let's dive in…
Heatmaps
JupyterLite includes the Plotly Python library, which is capable of visualizing data in a variety of ways. Heatmaps are one of the supported visualization methods. Heatmaps make it easy to spot outliers—imagine, for example, someone was experimenting with an expensive cluster and forgot to shut it down.
Aggregation
As mentioned previously, Resoto Notebook allows you to harness the power of pandas, a popular Python package for data analysis. The pandas DataFrame structure is a table-like object that allows for easy querying, filtering, and aggregation of data.
Let's try aggregating the number of cores in running instances per cloud, per region:
Of course, it does not end with heatmaps and aggregation! Since it is possible to execute arbitrary Python code in a notebook, the possibilities for data analysis are endless.
Once you're finished working with the notebook, it will be saved in your browser's local storage.
Usage
Jupyterlite support will be a part of Resoto 3.0. If you would like to try it out today, you can install the edge version of Resoto. Then, open https://<Resoto Core hostname or IP address>/notebook in your web browser. No need to install extra packages or run a Jupyter server!
How It Works
There are two main components: Pyodide and JupyterLite. JupyterLite is a distribution of JupyterLab that runs in a browser, and Pyodide is a port of CPython to WebAssembly/Emscripten.
JupyterLite
We integrated JupyterLite with Resoto by adding its static assets generation to our build pipeline.
During the asset generation process, JupyterLite creates a distribution for three components: JupyterLab, "classic" Jupyter, and an interactive REPL. The last two components are not required and we excluded them from the build process to reduce the artifact size. Additionally, we disabled sourcemaps generation. These measures allowed us to cut the size of the resulting distribution from 30 megabytes down to 9.7 megabytes.
It can be all configured via the launch flags when the JupyterLite build command is called. One convenient hack we found is to define the launch flags in the configuration file called jupyter_lite_config.json:
{
"LiteBuildConfig": {
"lite_dir": "jupyterlite_extras",
"output_dir": "resotocore/jupyterlite",
"no_sourcemaps": true,
"apps": [
"lab"
]
}
}
We also used a custom Pyodide version, defined in the jupyter-lite.json configuration file.
Pyodide
Pyodide is a port of CPython to WebAssembly/Emscripten. Pyodide allows you to easily run Python code in the browser. In the hierarchy of abstraction layers, Pyodide sits in between your code and the browser:
This, however, comes with some limitations. When we want to run a Python package that spawns subprocesses or opens sockets to make network calls, there is simply no such thing in the browser environment!
That posed a problem for us: we rely on a resotonotebook package which uses the requests library, and this library is opening sockets and making network calls. Fortunately, Pyodide provides a Python proxy for JavaScript APIs, which allows us to access the browser APIs using Python!
Let's look at Resoto Notebook for some examples:
import js
import pyodide
path = f"/graph/{graph}/search/list"
url = js.URL.new(path, js.location.origin)
options = {"method": "POST"}
options = pyodide.ffi.to_js(options, dict_converter=js.Object.fromEntries)
response = await js.fetch(url, options)
Here, we make a POST request using the fetch API in Python. JavaScript APIs are available in the imported js package. When we want to provide Python data structures to the JavaScript functions, Pyodide performs the type conversion implicitly. However, to transform a dictionary to a JavaScript object we need to use a ffi.to_js() Pyodide call (otherwise, the implicit conversion will produce a JavaScript Map).
To get JavaScript types back into Python world, we use the to_py() method:
json = await response.json()
return json.to_py()
async/await syntax is seamlessly interoperable in both Python and JavaScript thanks to Pyodide, which makes it easy to call into the JavaScript ecosystem. Additionally, JupyterLite has its own event loop, which makes it simple to await on asynchronous calls inside notebooks.
Final Thoughts
It is quite exciting to see that the current state of the web allows us to bring powerful tools directly into the browser. We hope that you will find browser-based notebooks a useful tool for exploring your cloud assets and getting insights.
