Applications might hang if underlying RBD image is deleted

[caroav]

If the RBD image that is behind an nvmeof ns is being deleted, while the ns is mounted to the host, it might cause IOs to hang. We should consider 2 things:

1. Can we add a warning in the RBD CLI to warn the user if he is about to delete an image that is used by a namespace? I think that the RBD can tell that by checking the metadata (nvmeof set something there). @idryomov?

2. Can the nvmeof GW get the event that the image was deleted, and automatically delete the namespace? There is no point in keeping this namespace, if the image was deleted. @gbregman

[idryomov]

1. Can we add a warning in the RBD CLI to warn the user if he is about to delete an image that is used by a namespace? I think that the RBD can tell that by checking the metadata (nvmeof set something there). @idryomov?

If the gateway has the image open (in read-write mode, at least), the user wouldn't be able to remove it, except in the case of a prolonged loss of connectivity between the gateway and the OSDs. Any image that is open in read-write mode has a watcher registered on it and that prevents the image from being deleted:

$ rbd rm test
2026-04-18T20:33:36.292+0000 7f9f1cb1d640 -1 librbd::image::PreRemoveRequest: 0x56107a564070 check_image_watchers: image has watchers - not removing
Removing image: 0% complete...failed.
rbd: error: image still has watchers
This means the image is still open or the client using it crashed. Try again after closing/unmapping it or waiting 30s for the crashed client to timeout.

[idryomov]

If the image is being deleted at the time the gateway attempts to open it, the open API would fail with ENOENT error.