[pve-devel] [PATCH v3 manager 1/2] HDEdit: check iothread by default and move it from advanced section

[Fabian Ebner]

Am 19.05.22 um 15:35 schrieb Matthias Heiserer:
> On 18.05.2022 11:40, Fabian Ebner wrote:
>> Am 12.05.22 um 11:24 schrieb Matthias Heiserer:
>>> Existing disks are not changed by this.
>>> Especially in benchmarks, iothreads significantly improve IO
>>> performance.
>>>
>>>
>>> Signed-off-by: Matthias Heiserer <m.heiserer at proxmox.com>
>>> ---
>>>
>>> Changes from v2:
>>> * also check iothread when adding a disk to an existing VM and
>>>   scsi single
>>> * use bind instead of hardcoded true
>>
>> This feels like to much automagic to me, because changes to the checkbox
>> (even if checkbox is for a virtio disk) will change the controller type
>> and vice versa. This also makes it impossible to only set iothread on
>> certain disks or use the "Virtio SCSI single" controller type without
>> setting iothread.
>>
>> Is it possible to instead have the checkbox be invalid with an
>> appropriate error for the user when it's a bad configuration?
> 
> Changes to the checkbox already change the Controller to Virtio SCSI
> (single), regardless of what was selected before. Anyways, the automatic
> change only happens in the wizard.
> 

You're right, and I'd argue that the current behavior isn't ideal either
;). I guess with only a single scsi disk it makes sense to automatically
switch, because the iothread setting would be invalid otherwise. It also
happens for a virtio disk though, where the scsi controller type isn't
even visible in the disk edit tab. One can still argue that it's just
not relevant there. But we switched to using a multi-disk panel some
time ago, and in that context it's just confusing, because changes to
each iothread checkbox will affect the scsi controller type.

> In case that a user wants iothread with a controller other than SCSI
> single or on a single disk, they can change that in the VM config, but I
> don't think that this is a common problem, at least much less common
> than wanting iothread enabled per default.
> 
> But I guess we can show a warning instead