Using Dask as a backend for joblib runner

[kforeman]

I'd like to use a Dask cluster on EMR to run a bunch of Hydra jobs in parallel. I have my experiments running locally using the standard joblib multi-run launcher, and I think next steps would looking something like this:

• Set the joblib launcher backend to dask
• Somewhere(?) create a Dask client that's connected to the cluster, as in the joblib Dask example

But that last part I'm not totally clear on where to do it. Inside my e.g. run.py before @hydra.main() maybe?

I'll give it a shot next week once I've had time to setup the Dask cluster for testing, but thought I'd check if anyone else has already figured this out. I'll happily write up my findings - searching for Hydra + Dask + joblib doesn't turn up anything currently.