Giters

burnettk / delete-docker-registry-image

If you are running a private v2 docker registry, and you are storing your data on disk, running this script from the machine where the data lives will let you fully delete an image or tag

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

deletion error

iaon opened this issue · comments

commented

./delete_docker_registry_image --image XXX/YYY:1.2.0-rc.1 -v

It works without error messages.

./delete_docker_registry_image --image XXX/YYY:1.1.0-rc.7 -v

And I see error

Traceback (most recent call last):
File "./delete_docker_registry_image", line 336, in
main()
File "./delete_docker_registry_image", line 325, in main
cleaner.delete_repository_tag(image, tag, force=args.force)
File "./delete_docker_registry_image", line 213, in delete_repository_tag
if self._layer_in_same_repo(repo, tag, layer):
File "./delete_docker_registry_image", line 165, in _layer_in_same_repo
layers = self._get_layers_from_blob(manifest)
File "./delete_docker_registry_image", line 85, in _get_layers_from_blob
return get_layers_from_blob(data_path)
File "./delete_docker_registry_image", line 38, in get_layers_from_blob
with open(path, "r") as blob:
IOError: [Errno 2] No such file or directory: '/mnt/registry-data/docker/registry/v2/blobs/sha256/9f/9fd95e11750dd5b53d8de7c2f5f77f122baf96fddba470f9085918fbb2bf5376/data'

Where 9fd95e11750dd5b53d8de7c2f5f77f122baf96fddba470f9085918fbb2bf5376 is a hash related to first tag 1.2.0-rc.1 -v

I think it happens because first command did't clean properly. And XX/YYY/_manifests/tags/1.2.0-rc.1 still not empty

root@ip-172-31-128-219:/mnt# grep -r 9fd95e11750dd5b53d8de7c2f5f77f122baf96fddba470f9085918fbb2bf5376 *
registry-data/docker/registry/v2/repositories/XXX/YYY/_manifests/tags/1.2.0-rc.1/current/link:sha256:9fd95e11750dd5b53d8de7c2f5f77f122baf96fddba470f9085918fbb2bf5376

Complete output can be found
https://gist.github.com/iaon/16e467e25b3c78234142

This issue exist only for python version of script, so I have to rollback to BASH revision.

commented

Thanks for the report.

As the author of the python rewrite, I will try to fix this issue and add test for it, and will appreciate any help.

commented

@abulimov i think the check at line 225 might be a causing the issue where @iaon sees the tag directory 1.2.0-rc.1 not getting fully deleted. the intent of checking if at_least_one_blob_was_used_elsewhere was a sort of extra precaution against the script corrupting data, and i assumed it wouldn't be a huge annoyance since generally at least one blob would in fact be used somewhere else. when the script noticed that it was deleting a bunch of blobs and absolutely none of them were used anywhere else, i guess i got a little scared and wondered if there might not be a bug in the script, therefore i didn't delete anything unless the user specified --force. in the python script, it's still deleting blobs/revisions/layers/etc in this case. in retrospect, this feature is confusing and i would be in favor of you removing it (all references to at_least_one_blob_was_used_elsewhere as well as the --force option) if you're willing.

so the missing unit tests might be when you're deleting a tag or a full repo (i think this bug would occur there, too) and nothing else is using any of the blobs being deleted.

commented

👍 Great!