Hmmm, interesting. That cluster set-up may not be well suited to any CPU-bound work you do subsequently. My guess is that you may be ending up with different environments across your scheduler and workers, but I’m not sure why this is happening for you.
I found your Intake-ESM datastore on gdata/zn44 and I am still unable to reproduce the error using your code and conda/analysis3-26.01.
What about any code from conda init? See here