Xen project Mailing List

On Tue, Nov 27, 2012 at 9:24 PM, Mike McClurg <mike.mcclurg@xxxxxxxxxx> wrote:

On 27/11/12 15:44, Rushikesh Jadhav wrote:
> Hi Mike and Andrei,
>
> Thanks for making the discussion open about this famous issue.
>
> Many times we face this issue. If a host is hang then the VMs from that
> host are unable to boot on any other host until the database VDI RW flag
> is reset. Usual error is "The VDI is already attached in RW mode". The
> solution procedure is to reboot the hang host or resetvdis manually.
>
> If we can do on-boot=reset then this will help to quickly boot failed
> VMs on new host without errors. As mentioned this will apply to start
> and restart operation but will "xe vm-reset-powerstate" be affected ?
>
> Regards,
> Rushikesh

Hi Rushikesh,

I don't think that on-boot=reset does what you think it does. Setting
on-boot=reset on a VDI will force it to "reset" on attach, meaning that
it will delete the leaf node of the VHD tree that makes up the VDI.

Basically, this is a feature which lets you create a VM which will roll
its disks back to a specified state every time you reboot. All of the
writes that you made to the disk before the reboot (and after setting
on-boot=reset) will be gone.

The problem is that there is a bug where this action happens on VM
migrate! This is what we're trying to fix.

Mike

Thank for the further info. So are you making a magic RO VDI which can hold the writes till its attached ?
That sounds good to me :)

Do you think that there is any usefulness to what I thought about on-boot=reset ?

> On Tue, Nov 27, 2012 at 5:45 PM, Andrei Lifchits

> <Andrei.Lifchits@xxxxxxxxxx <mailto:Andrei.Lifchits@xxxxxxxxxx>> wrote:
>
> Hi Mike,
>
> Sounds good to me. Should be pretty easy to change.
>
> Andrei
>
> ________________________________________
> From: Mike McClurg
> Sent: Monday, November 26, 2012 4:13 PM

> To: xen-api@xxxxxxxxxxxxx <mailto:xen-api@xxxxxxxxxxxxx>; Andrei

> Lifchits
> Subject: Move blktap2.reset_leaf call to new SR command "epoch_begin"
>
> Hi Andrei,
>
> (Note that I'm CC'ing the xen-api list to encourage more open
> development discussions.)
>
> I'm working on the VDI reset bug we talked about last week, where
> setting on-boot=reset on a VDI, and then migrating the VM, triggers the
> VDI reset after the migration is complete, because VDI reset is called
> on VDI attach.
>
> The fix for this should be to only call blktap2.reset_leaf when the VDI
> attach operation is performed during a VM start or reboot. There is a
> set of new SMAPIv2 calls in Xapi: begin_epoch, end_epoch, which are
> called by Xapi when a VM is started and stopped (respectively), and also
> during a VM reboot. These functions are not called during a VM migrate.
>
> My proposal is that we add two new SR commands, begin_epoch and
> end_epoch, and that we move the VDI reset code into begin_epoch. The two
> corresponding Xapi functions would then call these SR commands.
>
> Does this sound like a good solution to the migrate triggers VDI reset
> bug? Did I miss anything here? Thanks,
>
> Mike
>
> _______________________________________________
> Xen-api mailing list

> Xen-api@xxxxxxxxxxxxx <mailto:Xen-api@xxxxxxxxxxxxx>
> http://lists.xen.org/cgi-bin/mailman/listinfo/xen-api
>
>

Re: [Xen-API] Move blktap2.reset_leaf call to new SR command "epoch_begin"