Add labextension setup

[ScottWales]

Experiment adding dask-labextension, as suggested in #1

To try out you'll need to create a conda environment, as labextension isn't installed in analysis:

conda create -n proxytest python=3 nodejs jupyterlab dask-labextension -c conda-forge
conda activate proxytest
jupyter labextension install dask-labextension

then connect to that environment using ./gadi_jupyter -e proxytest

Things I like

• No need to forward the dask dashboard port
• Integration with the jupyterlab dashboard

Things I don't like

• Poor integration with the notebooks (there is a button to insert the connection URL, but presumably that changes each time you connect to Gadi)
• Would probably want to set up a central configuration file for most settings rather than set up dask-labextension through environment variables

[jmunroe]

dask-labextension can also do discovery of the dashboard of a client started within the notebook (click on the little magnifying glass icon)

And, yes, the recommendation would be to define .yml for default dask cluster creation. I'll look again dask_jobqueue with PBSCluster . I think that's the right way to go.

[ScottWales]

The autodiscovery doesn't seem to work without forwading the dashboard port as well, I've opened dask/dask-labextension#109 to see if it can be configured to use a proxy

[ScottWales]

This is working well now, the dashboard is being properly proxied. For the moment I think we'll stick with configuring using environment variables - most of the settings are dependent on the resources given to the PBS job