2:42 p.m.
Hi everyone,
I have found some issues with this patch:
http://gerrit.ovirt.org/#/c/18638/
It allows the user to disable the VirtIO-SCSI disk interface during the VM creation. The
problem is that the user still can add, attach and hotplug disks with the VirtIO-SCSI
interface type, but when the user does so, libvirt automatically creates a LSI Logic SCSI
controller and connects the new disk to it.
How can this problem be solved? Should the VirtIO-SCSI interface type be hidden from the
user in case it wasn't enabled, or should the engine enable the VirtIO-SCSI
controller, hotplug it, then hotplug the disk into it transparently?
Thanks,
Vitor de Lima
3:05 p.m.
Hi Vitor,
The new VirtIO-SCSI enabled checkbox is an indication whether to
attach a VirtIO-SCSI controller when running the VM.
It should be enabled automatically on cluster >= 3.3.
When disabled, I think it's preferable not to add a new controller
automatically when running the VM as it requires creating/attaching
a new VmDevice - which we refrain of on VmInfoBuilder flows
(and since it might be confusing to the user...).
As an alternative, I've planned to add a warning in the dialog
or create a canDo message to prevent running the VM at all.
I'm not sure we should hide the option from disk interfaces
list as it's already being filtered using VirtIoScsiEnabled
ConfigurationValue (and using OsInfo soon...).
Let me know what you think and thanks a lot for the input!
Daniel
----- Original Message -----
4:34 p.m.
Hi Daniel,
I asked this question because I have implemented a filter to show only compatible disk
interfaces (in change #17964). The main purpose of this patch is to hide the IDE interface
type when creating disks for PPC64 VMs (since IDE is not supported on this architecture).
If it was decided that the VirtIO-SCSI interface type should be hidden from the user in
case it was disabled, I would have to modify that patch a little bit.
Another issue is that in change #18622 the support for a PPC64-specific controller, the
SPAPR VSCSI controller, was introduced. But the code was created based on the assumption
that the VirtIO-SCSI controller was always present, and this isn't the case anymore.
And another patch that I will work on really soon will add support to create disks that
are connected to this interface.
So, I would like some feedback before changing these patches. Is a validation on the
backend enough to block the user from using an inexistent controller? Should the frontend
be changed as well? What would be a good approach to handle multiple SCSI controllers in a
VM (were the presence of one of them is optional)?
Thanks,
Vitor
5:23 p.m.
----- Original Message -----
That is not the case. Point is to allow users to add a scsi controller to a VM that does
not have any scsi disks. This is needed in case the user would later on like to hotplug
scsi disks. Without this, hotplug would fail since there is no scsi controller in the
VM.
When the VM is down, you can always attach a scsi disk and if there is no scsi controller
then engine will automatically add one.
If you have a need for a VM with more than 256 LUNs then I'd say that the easiest
behaviour to understand would be to hide the existence of the controllers from the user
altogether and just add support to automatically hotplug controllers on the fly when
existing controllers are reaching their limit.
7:31 a.m.
----- Original Message -----
For consistency, I'll filter the interface from the list when VirtIO-SCSI is
disabled.
So yeah, keep that in mind when modifying your patch.
Can you filter the option out or add a warning when VirtIO-SCSI is disabled?
You should block it in the engine on canDo and filter the option / warn about it in the
UI.
Regarding multiple controllers, are you referring to multiple types of controllers
or just multiple VirtIO-SCSI devices?
10:52 a.m.
Hi Daniel,
I was referring to multiple types of controllers. For now, I think it is enough to have
only one optional VirtIO-SCSI controller per VM, and (on the ppc64) one obligatory SPAPR
VSCSI controller. The main question is how to handle the addresses when the VirtIO-SCSI is
disabled and there is only the SPAPR VSCSI controller, and how to handle if the user
enables VirtIO-SCSI after the creation of the VM (I think the addresses for the existing
disks must remain stable, so the controller index must be 1 for the VirtIO-SCSI and 0 for
the SPAPR VSCSI in this case).
Do you have any tips on how to implement the disk/cdrom addressing in these cases?
Thanks,
Vitor
12:58 p.m.
----- Original Message -----
Regarding the controllers, the address of each controller should be kept
separately - i.e. saved as a VmDevice for maintaining a stable address.
Most changes should probably be done on VmDeviceUtils (look at updateVmDevices
and copyVmDevice methods) and VmInfoBuilder. Disk/cdrom devices should
be treated similarly.
As for changing the disk interface, we're clearing the device's address
when switching an interface; so I don't think there's an issue there.
More info on stable addresses can be found here:
http://www.ovirt.org/Features/Design/StableDeviceAddresses
4210
days inactive
4211
days old
6 comments
3 participants
participants (3)
-
Ayal Baron -
Daniel Erez -
Vitor de Lima