Matthew Rocklin

Dask should be robust to this. We shouldn't rely on users to be careful here. I think that someone needs to look deeply at how we currently count references and...

In other words. This isn't something that we want to work around or avoid. This is something that we should dive into. Diving into it is hard though and will...

Yeah, it's non-trivial. I would probably start by printing out or otherwise recording every time the client and scheduler think that the task has been created or cancelled. I think...

OK, so I should clarify my thinking from before. Certainly we need downstream deployment projects (dask-kubernetes, dask-cloudprovider, etc.) to continue functioning. What I intended to say before is that I...

First, exciting results. I'm excited. Also, kudos on the clear presentation of those results. If it is easy to share performance reports for the shuffling case I would be curious...

@gjoseph92 I'm curious, any luck digging into these and seeing what was going on?

@gjoseph92 does anything here sound familiar to you?

I think that we should not run the profiler during non-profiling tests. On Wed, Apr 6, 2022 at 3:54 AM Florian Jetter ***@***.***> wrote: > We constantly have a profiler...

It's not yet clear to me what `RemoteCluster` would provide. Client's don't need a Cluster object to operate, they just need a scheduler address. If the desire is to support...

The coarsen function might be relevant here http://dask.pydata.org/en/latest/array-api.html#dask.array.coarsen On Fri, May 25, 2018, 4:45 PM Matt McCormick wrote: > Stages: > > 1. Basic support for dask.array > 2. Stream...