rapidsai / jupyterlab-nvdashboard

A JupyterLab extension for displaying dashboards of GPU usage.
BSD 3-Clause "New" or "Revised" License
598 stars 78 forks source link

Upgrading to Bokeh 2 #66

Closed jakirkham closed 4 years ago

jakirkham commented 4 years ago

Current we are pinning jupyterlab-nvdashboard to bokeh version 1. Would be good to fix any remaining compatibility issues so bokeh version 2 can be used.

jacobtomlinson commented 4 years ago

Agreed. I think Bokeh 2.1 has had a bunch of issues, so I'm tempted to hold off until their next patch release.

jakirkham commented 4 years ago

Do you happen to recall the issues?

owenlamont commented 4 years ago

I was just about to request this too... would be nice to install nvdashboard in Python environments that already have Bokeh 2+. I don't know how much work is involved in this but happy to try and help when you think it's good to go ahead.

owenlamont commented 4 years ago

One issue we have with Bokeh 2 on conda is if a user tries to install nvdashboard into a Python environment that already has Bokeh 2 conda won't force them to downgrade Bokeh - they will just get an older version of nvdashboard that pre-dates the Bokeh<2 requirement which will be broken in their environment. I'm not sure if pip has the same issue... I didn't test that. I don't know if there is a way to remove the pre 0.3 recipes from conda-forge @jakirkham

jakirkham commented 4 years ago

Probably the thing to do is hot-fix older jupyterlab-nvdashboard packages to have this constraint as well. For example here's how we addressed a similar issue with distributed ( https://github.com/conda-forge/conda-forge-repodata-patches-feedstock/pull/32 ).

jakirkham commented 4 years ago

FWIW Bokeh 2.1.1. shipped today. Not sure if that includes fixes to the issues observed, but may be worth checking 🙂

owenlamont commented 4 years ago

Probably the thing to do is hot-fix older jupyterlab-nvdashboard packages to have this constraint as well. For example here's how we addressed a similar issue with distributed ( conda-forge/conda-forge-repodata-patches-feedstock#32 ).

Good to know. I'll try to look into making the patches this coming weekend for conda-forge.

AjayThorve commented 4 years ago

+1 for this. I just finished upgrading cuxfilter to bokeh 2.1.1, and the issues were fairly straight forward to fix. We even have @bryevdv on the viz team now, so let us know if we can help getting this integrated

jacobtomlinson commented 4 years ago

Yeah 2.1.1 is the release I was waiting for. Thanks for testing things @AjayThorve. I'll open a PR to update the version.