Re: [pve-devel] [PATCH v4 serries 0/4] disk reassign: add new feature

[Aaron Lauterer <a.lauterer@proxmox.com>]

Does anyone have time to take a look at the latest iteration of these patches?

Thx :)

On 10/2/20 10:23 AM, Aaron Lauterer wrote:
> This series implements a new feature which allows users to easily
> reassign disks between VMs. Currently this is only possible with one of
> the following manual steps:
> 
> * rename the disk image/file and do a `qm rescan`
> * configure the disk manually and use the old image name, having an
>      image for VM A assigned to VM B
> 
> The latter can cause unexpected behavior because PVE expects that the
> VMID in a disk name always corresponds to the VM it is assigned to. Thus
> when a disk, original from VM A was manually configured as disk for VM B
> it happens that, when deleting VM A, the disk in question will be
> deleted as well because it still had the VMID of VM A in it's name.
> 
> To issue a reassign from the CLI run:
> 
> qm reassign_disk <source VMID> <target VMID> <drive key>
> 
> where <drive key> is the config key of the disk, e.g. ide0, scsi1 and so
> on.
> 
> The following storage types are implemented at the moment:
> * dir based ones
>      * directory
>      * NFS
>      * CIFS
>      * gluster
> * ZFS
> * (thin) LVM
> * Ceph RBD
> 
> v3 -> v4:
> * revert intermediate storage plugin for directory based plugins
> * add a `die "not supported"` method in Plugin.pm
> * dir based plugins now call the file_reassign_volume method in
>    Plugin.pm as the generic file/directory based method
> * restored old `volume_has_feature` method in Plugin.pm and override it
>    in directory based plugins to check against the new `reassign` feature
>    (not too happy about the repetition for each plugin)
> * task description mapping has been moved from widget-toolkit to
>    pve-manager/utils
> 
> 
> v2 -> v3:
> * change locking approach
> * add more checks
> * add intermedia storage plugin for directory based plugins
> * use feature flags
> * split up the reassign method to have a dedicated method for the
>      renaming itself
> * handle linked clones
> * clean up if disk used to be replicated
> 
> I hope I didn't forget anything major.
> 
> v1 -> v2:
> print info about the new disk volid and key at the end of the job so it
> shows up in the CLI output and task log
> 
> Changes from RFC -> V1:
> * support to reassign unused disks
> * digest for target vm config
> * reorder the checks a bit
> * adding another one to check if the given key for the disk even exists
>    in the config.
> 
> qemu-server: Aaron Lauterer (2):
>    disk reassign: add API endpoint
>    cli: disk reassign: add reassign_disk to qm command
> 
>   PVE/API2/Qemu.pm        | 156 ++++++++++++++++++++++++++++++++++++++++
>   PVE/CLI/qm.pm           |   2 +
>   PVE/QemuServer/Drive.pm |   4 ++
>   3 files changed, 162 insertions(+)
> 
> 
> 
> storage: Aaron Lauterer (1):
>    add disk reassign feature
> 
>   PVE/Storage.pm                 | 15 ++++++++--
>   PVE/Storage/CIFSPlugin.pm      | 19 +++++++++++++
>   PVE/Storage/DirPlugin.pm       | 19 +++++++++++++
>   PVE/Storage/GlusterfsPlugin.pm | 19 +++++++++++++
>   PVE/Storage/LVMPlugin.pm       | 24 ++++++++++++++++
>   PVE/Storage/LvmThinPlugin.pm   |  1 +
>   PVE/Storage/NFSPlugin.pm       | 19 +++++++++++++
>   PVE/Storage/Plugin.pm          | 51 ++++++++++++++++++++++++++++++++++
>   PVE/Storage/RBDPlugin.pm       | 31 +++++++++++++++++++++
>   PVE/Storage/ZFSPoolPlugin.pm   | 28 +++++++++++++++++++
>   10 files changed, 224 insertions(+), 2 deletions(-)
> 
> manager: Aaron Lauterer (1):
>    ui: tasks: add qmreassign task description
> 
>   www/manager6/Utils.js | 3 +++
>   1 file changed, 3 insertions(+)
> 
>