Issue with parallelization in sea ice notebook?

[mnlevy1981]

Describe the bug
@dabail10 has been reporting occasional RuntimeError: NetCDF: Not a valid ID errors when running a notebook in parallel. I found a two-year-old discussion from Australia, but their fix was to run with a single thread per worker and we already set threads_per_worker=1 when creating the LocalCluster. This is verified in the client object, where there are 16 processes and 16 threads:

<Client: 'tcp://127.0.0.1:39305' processes=16 threads=16, memory=120.00 GiB>

To Reproduce
On casper or derecho, run cupid-diagnostics --ice on a compute node with multiple cores; sometimes the Hemis_seaice_visual_compare_contour.ipynb fails.

Expected behavior
This error should not crop up.

Additional context
There's additional discussion at pydata/xarray#7079 (I found that issue because it is mentioned in the Australian discussion), but it all seems to claim the issue is limited to threads_per_worker>1 so maybe it's not related to Dave's problem?

Running cupid-diagnostics --ice --serial will avoid the issue, but it will also run slower because it won't be parallelized.