Remove wheels that are older than a threshold time limit

[bsipocz]

There are some 1 and 2 month old numpy wheels that haven't been identified for removal. They all have rc1 in their version number, so that may trip up the old wheel removing GHA.

https://anaconda.org/scientific-python-nightly-wheels/numpy/files?page=2

[matthewfeickert]

There are some 1 and 2 month old numpy wheels that haven't been identified for removal.

@seberg as your credentials were used to upload these can you remove them if you think that aren't useful currently? I'd prefer for project maintainers to do this.

They all have rc1 in their version number, so that may trip up the old wheel removing GHA.

No, the name information is not used beyond the project.

The reason those still exist is that they have been uploading numpy-2.0.0.dev0 wheels that overwrite the previous one and so they still only have 5 wheels on their project — the current limit.

upload-nightly-action/.github/workflows/remove-wheels.yml

Lines 18 to 19 in a2ba178

	env:
	N_LATEST_UPLOADS: 5

upload-nightly-action/.github/workflows/remove-wheels.yml

Lines 68 to 70 in a2ba178

	anaconda show "${ANACONDA_USER}/${package_name}" &> >(grep '+') | \
	awk '{print $2}' | \
	head --lines "-${N_LATEST_UPLOADS}" > remove-package-versions.txt

So this isn't a bug in the current code, but a new feature request to also remove wheels past a certain time threshold. Seems like a good idea.

[seberg]

I have deleted all but the current ones.

EDIT: To be fair, I somewhat liked the exact version in the name with an auto-deleter of old ones, whatever the exact threshold. But numpy switched from a new version for each wheel to replacing the same version.

[matthewfeickert]

Thanks @seberg!

[bsipocz]

Ahh, OK. This case, we should maybe have another, time-limit logic in the cleanup script, too. E.g. delete everything that is ~month old? At that point I think we can argue that it's not really a nightly build, but something broken.

[matthewfeickert]

Yeah 1 month seems very reasonable. 👍

[bsipocz]

hah, I see that you already modified the issue here to reflect that. Thanks!

[matthewfeickert]

Asked about how much datetime metadata we can get from Anaconda Cloud easily over on anaconda/anaconda-client#682.