distributed writes

[d-v-b]

I'm trying to write a tensorstore array using dask on a single machine. I have a bunch of worker processes that each a) call tensorstore.open(spec, write=True).result(), and then attempt to assign a numpy array to a contiguous, chunk-aligned range of indices in the resulting tensorstore array. This does not appear to be running in parallel.

How can I make this work? Are there any best practices for writing to tensorstore arrays from multiple processes that I should be aware of?

[jbms]

This should work. Can you send a simple example that demonstrates the problem so that I can look into it?

[d-v-b]

I made a minimal reproducible example which confirms what you say -- multiprocess writes work fine. So I must be doing something wrong in the code that wasn't parallelizing properly.