Jupyter + DAR compatibility exploration for running Stencila on binder
This project is part of the eLife Innovation Sprint 2018 and Mozilla Global Sprint 2018 (see mozilla/global-sprint#317)
This project comprises two modules:
- a JavaScript package for installing and running Stencila (client/UI and services) in a Jupyter container
- a Python package for installing and running a proxy (based on
nbserverproxy) to access the UI as well as services provided by Stencila; the package also extends the Jupyter UI to add a "New Stencila Session" button
We first used Stencila's development build to run the app using node make -w -s -d /our/own/dir, but struggled a bit to configure the file storage, i.e. the dar-server, to use the directory we want to, and to run it in a full path configured by us instead of make.js starting the dar-server relative to __dirname.
Eventually we ended up implementing our own minimal npm package that pulls in Stencila as a dependency and runs the dar-server and static file server for the app using the files from the dist directory.
See the file stencila.js for details.
This gives us control of the paths and let's us get rid of complex development features (substance-bundler etc.).
We also made our own version of app.js, getting rid of the virtual file storage stuff (vfs), defaulting storage to fs (file system), because that is what is needed for Jupyter - we do not need to host any examples.
In the same line, we built own index.html (based on example.html) and serve that, which allows us to directly render a DAR document instead of a listing of examples and instruction and to use our app.js.
Relevant path configurations comprise the local storage path as well as the URLs used by the client, accessing the dar-server through the nbserverproxy.
The Dockerfile installs our helper npm package and adds + configures the nbserverproxy tool (see requirements.txt and jupyter_notebook_config.py).
Stencila has "execution contexts" (the equivalent of Jupyter's "kernels") for R, Python, SQL, Javascript (in the browser), and Node.js. Execution contexts differ from kernels in a number of ways including local execution and dependency analysis of cells. Both of these are necessary for the reactive, functional execution model of Stencila Articles and Sheets.
We could install these execution contexts in the Docker image. However, Stencila also has a JupyterContext which acts as a bridge between Stencila's API and Jupyter kernels. So, since the base jupyter/minimal-notebook image already has a Jupyter kernel for Python installed it we decided to use that. This does mean however, that some of the reactive aspects of the Stencila UI won't work as expected. Also the JupyterContext is not well developed or tested.
We have included the stencila-node Node.js package in the Docker image which provides the JupyterContext as well as a NodeContext (for executing Javascript) and a SqliteContext (for executing SQL) .
nbstencilaproxy provides Jupyter server and notebook extensions to proxy Stencila. It is based on nbrsessionproxy but does not include the support nbrsessionproxy has for JupyterLab.
Install
Install package:
pip install git+https://github.com/nuest/nbstencilaproxy
Install the extensions for all users on the system:
jupyter serverextension enable --py --sys-prefix nbstencilaproxy
jupyter nbextension install --py --sys-prefix nbstencilaproxy
jupyter nbextension enable --py --sys-prefix nbstencilaproxy
The Dockerfile contains an example installation on top of jupyter/r-notebook.
- Build the Docker image:
docker build --tag jupyter-dar .- Run the Docker image:
docker run -p 8888:8888 jupyter-dar-
Login by visiting the tokenized URL displayed e.g.
http://localhost:8888/?token=99a7bc13... -
Go to http://localhost:8888/stencila/ or click on the "New > Stencila Session" button on the Jupyter start page
BSD 3-Clause License