Comment 0 for bug 1511061

Revision history for this message
Hemanth Makkapati (hemanth-makkapati) wrote :

[0] shows a sample image that was left in an inconsistent state when a call to registry failed during image deletion.

Glance API makes two registry calls when deleting an image.
The first call [1] is made to to set the status of an image to deleted/pending_delete.
And, the other [2], to delete the rest of the metadata, which sets 'deleted_at' and 'deleted' fields in the db.

If the first call fails, the image deletion request fails and the image is left intact in it's previous status.
However, if the first call succeeds and the second one fails, the image is left in an inconsistent status where it's status is set to pending_delete/deleted but it's 'deleted_at' and 'deleted' fields are not set.

If delayed delete is turned on, these images are never collected by the scrubber as they won't appear as deleted images because their deleted field is not set. So, these images will continue to occupy storage in the backend.
Also, further attempts at deleting these images will fail with a 404 because the status is already set to pending_delete/deleted.

[0] http://paste.openstack.org/show/477577/
[1]: https://github.com/openstack/glance/blob/master/glance/api/v1/images.py#L1115-L1116
[2]: https://github.com/openstack/glance/blob/master/glance/api/v1/images.py#L1132