qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Qemu-devel] Re: [PATCH 04/22] savevm: do_loadvm(): Always resume the VM


From: Juan Quintela
Subject: [Qemu-devel] Re: [PATCH 04/22] savevm: do_loadvm(): Always resume the VM
Date: Wed, 21 Apr 2010 17:45:23 +0200
User-agent: Gnus/5.13 (Gnus v5.13) Emacs/23.1 (gnu/linux)

Luiz Capitulino <address@hidden> wrote:
> On Wed, 21 Apr 2010 15:28:16 +0200
> Kevin Wolf <address@hidden> wrote:
>
>> Am 20.04.2010 23:09, schrieb Luiz Capitulino:
>> > do_loadvm(), which implements the 'loadvm' Monitor command, pauses
>> > the emulation to load the saved VM, however it will only resume
>> > it if the loading succeeds.
>> > 
>> > In other words, if the user issues 'loadvm' and it fails, the
>> > end result will be an error message and a paused VM.
>> > 
>> > This seems an undesirable side effect to me because, most of the
>> > time, if a Monitor command fails the best thing we can do is to
>> > leave the VM as it were before the command was executed.
>> 
>> I completely agree with Juan here, this is broken.
>
>  Yeah, it's an RFC. I decided to send it as is because I needed feedback as
> this series is a snowball.
>
>> If you could leave the VM as it was before, just like you describe
>> above, everything would be okay. But in fact you can't tell where the
>> error occurred. You may still have the old state; or you may have loaded
>> the snapshot on one disk, but not on the other one; or you may have
>> loaded snapshots for all disks, but only half of the devices.
>> 
>> We must not run a machine in such an unknown state. I'd even go further
>> and say that after a failed loadvm we must even prevent that a user uses
>> the cont command to resume manually.
>
>  Maybe 'info status' should have a specific status for this too.
>
>  (Assuming we don't want to radically call exit(1)).

I tried a variation of this in the past, and was not a clear agreement.

Basically, after a working migration to other host, you don't want to
allow "cont" on the source node (it target has ever changed anything, it
would give disk corruption).

But my suggestion to disable "cont" after that got complains that people
wanted a "I_know_what_I_am_doing_cont". (not the real syntax).  Perhaps
it is time to revise this issue?

Later, Juan.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]