Updates needed to remove `DeprecationWarning`s

[padraic-shafer]

Some housekeeping is needed to respond to multiple DeprecationWarnings that are raised during testing.

• DeprecationWarning: on_event is deprecated, use lifespan event handlers instead.
  • Read more about it in the FastAPI docs for Lifespan Events.
  • I will submit a PR for this.
• DeprecationWarning: product is deprecated as of NumPy 1.25.0, and will be removed in NumPy 2.0. Please use prod instead.
  • I will submit a PR for this.
• DataFrame._data is deprecated and will be removed in a future version. Use public APIs instead.
  • This comes from the cachey dependency. It will likely require an upstream PR.

[padraic-shafer]

Submitted PR #676 for the FastAPI lifespan events

[padraic-shafer]

Submitted PR #677 to use numpy.prod() rather than numpy.product().

[padraic-shafer]

• DataFrame._data is deprecated and will be removed in a future version. Use public APIs instead.
  • This comes from the cachey dependency. It will likely require an upstream PR.

It looks like Pandas wants to remove its BlockManager. The roadmap is not yet clear on what the replacement will look like, and there is not suggested alternate for accessing the blocks. This access has remained stable for several years, and discussions suggest there is no hurry to change that.

So our best option here is probably to just ignore this warning. We could specifically filter it with action: ignore:DataFrame._data is deprecated and will be removed in a future version. Use public APIs instead.:DeprecationWarning.

References

• https://pandas.pydata.org/pandas-docs/version/0.25.2/development/roadmap.html?highlight=blocks#block-manager-rewrite
• DEPR: .as_blocks and .blocks pandas-dev/pandas#17302

[padraic-shafer]

Submitted PR #682 to ignore the warnings from cachey.nbytes().

[danielballan]

Maybe we can solve the cachey problem by simply dropping the dependency.

[DiamondJoseph]

Are all of these issues now resolved, can this ticket be closed?

[padraic-shafer]

For posterity, here is a checklist of PRs corresponding to the sub-issues raised in this issue:

• App uses lifespan context manager for startup + shutdown #676
• Replace numpy.product() with .prod() #677
• Ignore DeprecationWarning from Pandas via cachey #682
• Remove "object cache" misfeature; add simpler "resource cache" #708

This issue can now be closed.