When there are image chains problems, sometimes it is not possible to
determine the correct chain due to lost information like rotated logs. For
this cases, it is usually desirable to remove a volume from the chain but
not from the storage until the user confirms the VM is healthy and contains
the correct data. Currently this is done by moving files (file storage) or
renaming LVM tags (block storage) so VDSM does not find the particular
volume to be part of the chain.
Ideally I would like to have this done by a vdsm-tool talking to VDSM api.
The tool would be able to mark the volume to be ignored or not, list all
ignored volumes and perhaps even delete them (to free space).
How would you recommend proceeding with such tool? Some ideas:
- Have a reserved image UUID which contains all ignored volumes, so they
are not part of any chain.
- Add an "ignore" metadata attribute to each volume.
I like more the first option, but looking for opinions and ideas.
Show replies by date