Bokeh 3.1 support in Dask

Bokeh maintainer here.

I know that dask is pinning to bokeh<3 currently due to problems caused by the significant changes between bokeh 2.4 and 3.0. I think those problems are now dealt with and I’d like to find a way forward to remove that version pin.

Bokeh 3.1.0 RC is out now and the full release should occur in the next week or two. If I use the latest dask and distributed commits, increase distributed’s MAX_BOKEH_VERSION and use the Bokeh RC via conda install -c "bokeh/label/dev" bokeh then all of the dask-labextension possibilities seem to work fine in jupyter lab for me.

If I roll back to bokeh 3.0.3, the latest release on the 3.0 branch, then most of the dask dashboard works fine for me except for the task stream. We could probably identify the fixes in 3.1 that we’d need to backport to the 3.0 branch if we wanted to guarantee that the latest 3.0.* release works with dask.

I am wondering what the best way forward here is. Firstly maybe PRs for both dask and distributed which use the bokeh 3.1 RC in CI for full testing? Then there is the question of whether you are happy to permanently exclude bokeh 3.0.* from your dependencies or if release of a 3.0.4 is preferred?

Hi @ianthomas23, welcome here!

I’m not sure if this forum is the best place to discuss this, as core developers of Dask are somehow more focused on Github. I’ll ping some people anyway, but if you don’t get a response fast, I think you should consider opening a Github issue instead.

cc @mrocklin @jrbourbeau.

@jrbourbeau will know the right answer here. I think that he is managing this.