Kevin Paul

@jrhemstad I agree that we shouldn't limit ourselves to a single MPI implementation, but you also point out a problem that finding a solution that works for all implementations may...

Okay. I was unavailable the last couple of days to respond, so I recognize that I'm coming back to this conversation a bit late. I wanted to add some thoughts...

I agree with @jrhemstad, here. It is unfortunate that the MPI standard is...well...not very standard. However, this is the only appropriate way forward. We need to stick to the MPI...

I feel like some context was lost in the transfer of this issue to `dask-mpi`...? Not sure what exactly is being asked. @rcthomas I hate to ask you to repeat...

Is this related to #9 in any way?

Ah! Yes. This would be very cool. ...Let's talk about what this would look like, then, because I'm actually wondering if this is more of a fit for Dask-Jobqueue.

@rcthomas Is there a difference between the two approaches, other than that one uses a batch script and the other doesn't? In other words, would a user notice a difference...

@rcthomas Yes. I think you are correct that this doesn't fit into the Dask-Jobqueue paradigm. This cluster object should probably be defined in Dask-MPI using the `subprocess` algorithm like the...

@rcthomas Ok. A question for you about what you actually need. The command that you quoted above. Is this what you want to run when you click on the Dask...

@rcthomas No worries about getting busing with other things! I completely understand. The fact that `salloc` does not return immediately is definitely a complication, but I think there's a way...