Benjamin Zaitlen

So we use pynvml in two places: 1) getting the number of GPUs in machine -- this is easy to do 2) getting the CPU affinity for GPUs -- this...

Alternatively, if there is only one GPU on a jetson, does device affinity do anything ?

Should we close this if it is superseded by https://github.com/rapidsai/cudf/pull/12106 ?

I agree with @pentschev though I'm curious what version of CuPy you are using. When I perform the same commands above I see something quite different (using cupy 11.6 /...

> /tmp/tmpt76qy7aa' This does not look like a valid CuPy version ? How are folks configuring LD_LIBRARY_PATH ? That would explain why there are things like `/company/software/rhel8/gcc-10.2.0/lib64/libgomp.so` showing up. It's...

I believe @madsbk and @jakirkham did a fair amount of work around serialization generally and specifically for HLGs. If we relaxed the requirement for say `1)` how would this make...

There are errors in [test_spill_to_disk](https://github.com/dask/distributed/pull/4077/checks?check_run_id=1028278817#step:8:250) which seem related > lengths = struct.unpack_from(f"{n_frames}{fmt}", b, fmt_size) E struct.error: total struct size too long

How hard would it be to write a test for this where scheduler/worker output is redirected to a file and we set `--log-file` just as you do in the toy...

@jacobtomlinson if you have a moment would you be able to look this over ?

I'm sorry, I don't see a PR/Link/Repo for MPI4Dask. Can you please add this ?