pangeo-forge / pangeo-forge-bakery-images

Infrastructure and Image definitions for pangeo-forge bakeries
Apache License 2.0
0 stars 0 forks source link

Pin fsspec gcsfs s3fs to 2021.11.0 in 0.6.1 image #25

Closed cisaacstern closed 2 years ago

cisaacstern commented 2 years ago

What I am changing

The 0.6.1 image did not declare explicit versions for fsspec, gcsfs, and s3fs. As a result, it ended up with incompatible versions of fsspec and gcsfs, resulting in failed GCS writes discovered via https://github.com/pangeo-forge/pangeo-forge-gcs-bakery/issues/19.

As noted in https://github.com/pangeo-forge/pangeo-forge-recipes/pull/247, pangeo-forge-recipes 0.6.1 does not work with fsspec > 2021.11.0. To get a compatible complement of fsspec + gcsfs + s3fs, I have therefore pinned all three of these packages to 2021.11.0.

How I did it

Declared explicit versions for these packages in this image's requirements.txt.

How you can test it

This was blocking https://github.com/pangeo-forge/pangeo-forge-gcs-bakery/issues/19, and the original image was non-functional for interface with GCS, so I've already pushed this update to Docker Hub. Once we have more services relying on these images, it will probably be best to define some approval/testing process before updating them, but in this case, I just went for it. Hopefully this did not impact any one else's work.

cisaacstern commented 2 years ago

@sharkinsspatial, it looks like

https://github.com/pangeo-forge/pangeo-forge-bakery-images/blob/62aa27bf9f4bbb2c61f3ffe56140463bfe698dcc/images/pangeonotebook-2021.07.17_prefect-0.14.22_pangeoforgerecipes-0.5.0/Dockerfile#L11

and

https://github.com/pangeo-forge/pangeo-forge-bakery-images/blob/62aa27bf9f4bbb2c61f3ffe56140463bfe698dcc/images/pangeonotebook-2021.07.17_prefect-0.14.22_pangeoforgerecipes-0.6.0/Dockerfile#L11

are why the tests are failing, because presumably that repo and/or branch of fsspec no longer exists. Should we:

  1. Just remove these two image directories from this repo?
  2. Figure out what released version of fsspec they are compatible with, and make a separate PR to fix that?
  3. Not bother fixing this in any way, maybe just merge with the failing test, because we're going to work on the automated release refactor soon anyway?
  4. Something else

?

sharkinsspatial commented 2 years ago

@cisaacstern I would lean towards option 3 here.

cisaacstern commented 2 years ago

@cisaacstern I would lean towards option 3 here.

Works for me!