Run a previous version of Dask on Pangeo Cloud?

[paigem]

I am no longer able to run a piece of code using the package Rechunker on Pangeo Cloud that I have been able to up until about 5 days ago. I get an AssertionError, when running rechunker.rechunk() - similar to what is reported here. I am using an old (working) version of rechunker: 0.3.3.

I am guessing that the issue arises from:

1. The docker image update that happened recently and appears to have updated from Dask 2022.1.1 to Dask 2022.2.1.
2. Rechunker version 0.3.3 is not compatible with Dask 2022.2.1. (See this issue)

My question: is there a way to run the previous version of Dask (2022.1.1) on Pangeo Cloud at the moment so that I can keep running my Rechunker code?

[sgibson91]

I think the image that Pangeo Cloud is using would have to be rolled back for everyone

[paigem]

Ok, that’s not a great option then. Thanks for your response @sgibson91!

[rabernat]

Just chiming in quickly to say

• Sorry Paige! I know this is super annoying.
• I am fixing rechunker (see Refactor executors again by rabernat · Pull Request #112 · pangeo-data/rechunker · GitHub)
• From a broader perspective, in order to both be able to innovate quickly but also keep environments stable for users like Paige, our hubs need the ability to let users easily select any of the docker images when you log in. This is something we are working on with 2i2c

[rabernat]

If you want to try my fix for rechunker, please just pip install git+https://github.com/rabernat/rechunker.git@refactor-executors-again. (Make sure to also install it on the workers.)