Re: [pve-devel] [PATCH v2 manager 1/3] ui: restore: disallow empty storage selection if it wouldn't work

[Fabian Ebner <f.ebner@proxmox.com>]

Am 27.04.22 um 09:05 schrieb Thomas Lamprecht:
> On 25.04.22 09:28, Fabian Ebner wrote:
>> Am 23.04.22 um 11:38 schrieb Thomas Lamprecht:
>>> On 21.04.22 13:26, Fabian Ebner wrote:
>>>> Namely, if there is a storage in the backup configuration that's not
>>>> available on the current node.
>>>
>>> Better than the status quo, but in the long run all the "force all volumes to a single storage"
>>> on restore and also migrate isn't ideal for the case where one or more storages do not exist on
>>> the target node. An per-volume override would be nicer, but may require some gui adaptions to
>>> present that in a sensible way with good UX.
>>>
>>
>> In the UI, it could simply be part of the disk grid (proposed in patch
>> manager 3/3), only showing up for drives selected from the backup?
> 
> exactly what I thought too.

My first attempt using a widgetcolumn with a storage selector failed, as
it would issue an API call for each disk...I'll try to come up with some
way of sharing/fixing the store to make it work. In v3, I forgot to
group the override settings in a fieldset, will do so in v4.

> 
>>
>> In the back-end for migration, we have a storage-storage map, but here
>> we'd need a drive-storage map. It'd be possible to extend the 'storage'
>> parameter for the create/restore API call to be such a map, but I wonder
>> if going for a 'restore-drives' parameter being such a map (and
>> replacing the proposed 'preserve-drives' parameter) would be better?
> 
> hmm, possibly
> 
>>
>> The downside is, we'd have to choose between
>> A) preserve disk and config
>> B) preserve disk as unused
>> for the drives that are not present in the backup. A) would be more
>> convenient in the partial restore context, but B) is the current
>> default. Thus we need to keep B) if 'restore-drives' is not specified at
>> all for backwards-compatibility, but can choose A) if 'restore-drives'
>> is specified. But doing so seems a little inconsistent regarding user
>> expectation.
> 
> would a more general src:dest map help, for example (just to give the very
> rough direction meaning here):
> 
> not present or `scsi1:backup` <- would be restored as in the backup (config)
> `scsi1:store=foo` <- as in config put on another storage
> `scsi1:preserve` <- preserve from existing installation being overwritten
> 
> The actual left/right hand sides would need to get fleshed out to fit our
> use cases best, but 

I sent a v3 yesterday with something similar.