Remove pins on specific (and old) versions of pytz and fsspec

[DManowitz]

The latest version of this package (currently 2.7.1) has dependencies on pytz==2023.3 and fsspec==2023.6.0. It is currently mid-August 2024. Please verify that this package will work with more recent versions of these packages and unpin them from these specfic, old versions.

[andrii-i]

Hi @DManowitz. Thank you for creating this issue. This request makes sense and we are planning to update pytz and fsspec version dependencies at some point but this is not a priority / we are not working on it now. At the same time, could you please share more about why / if Jupyter Scheduler not using latest versions of pytz and fsspec is a problem / is a blocker for you?

[mbuechner]

Hi @andrii-i! I would upvote for this issue, because I have to downgrade my environment.

- fsspec             2024.10.0
+ fsspec              2023.6.0
- pytz                  2024.1
+ pytz                  2023.3
- psutil                 6.1.0
+ psutil                 5.9.8

Updating these dependencies would help ensure smoother integration with other libraries and reduce potential future issues, as projects relying on Jupyter Scheduler may encounter compatibility constraints if they require newer versions. So while it’s not an immediate blocker, keeping these dependencies up-to-date supports the stability and reliability of the Scheduler in the long term.

[DManowitz]

Sorry, I have been busy and not responding to this has not been my highest priority, but @mbuechner said exactly what I would say.

[andrii-i]

Hi @mbuechner @DManowitz. Thank you for the feedback. I'm looking into updating pytz and fsspec dependencies to the latest version.

[dlqqq]

Andrii discussed this issue with our team. This issue is challenging to solve because these packages don't follow semver, therefore we cannot know which versions work with Jupyter Scheduler unless we include them manually then test.

We've identified a short-term and long-term solution to this issue:

• Short-term solution: For pytz and fsspec, find the latest version of each that works, then raise the version ceiling to that version. This assumes that if versions X & Y work, then all versions between X & Y also work. This isn’t technically true (since any release could be breaking), but it is the most sensible heuristic that we can apply to fix this issue quickly for users.
• Long-term solution: Implement automated tests for each component that uses fsspec and pytz, then test each version within our specified version ranges.

[andrii-i]

Jupyter Scheduler 2.10.0 that includes support for latest versions of fsspec and pytz is released on PyPI and conda-forge 🚀