Matthew Rocklin

> Am I right in thinking that using dask-cuda-worker mostly makes sense when your job books at least two GPUs? Yes > Also are all the command-line arguments of dask-worker...

> when the job starts on a node, it finds an unused GPU (looking at lock files) and sets CUDA_VISIBLE_DEVICES That sounds like a nice solution. Do we know if...

No strong preference. What do you think is best? I would probably default to `unittest` just because that's what I've done before. That's not a very good reason on it's...

Given that we will be using pytest we can probably at least avoid the self.assertFoo methods :) On Fri, Mar 2, 2018 at 11:07 AM, jakirkham wrote: > Sure, using...

My original suggestion was to create a `unittest.TestCase` subclass that contains a set of tests, but that needs a bit of specialization to each cluster type. ```python class TestBasic(unittest.TestCase): Cluster...

I'm not sure how much reuse there is though. This might or might not make sense.

Bokeh released a new version yesterday that slightly breaks things with dask/distributed. I plan to issue a micro release to resolve the issue sometime soon.

https://github.com/dask/distributed/pull/803

I don't know of anything. If anyone has something a PR would be most welcome. On Thu, Dec 28, 2017 at 1:30 PM, swamidass wrote: > Any progress on this?...

Ah, my apologies, I see now that this is a PR. It looks like this PR needs help with testing and coverage as mentioned by @stumitchell above. If you have...