[Qemu-devel] Re: [PATCH] fix broken migration

[Glauber Costa <glommer@redhat.com>]

On Sat, Jul 25, 2009 at 03:37:24AM +0200, Paolo Bonzini wrote:
> On 07/24/2009 10:20 PM, Glauber Costa wrote:
>> While fixing migration with -S, commit
>> 89befdd1a6b18215153b8976682d57b7d03d5782 broke the rest of us.
>
> With this patch a migrated virtual machine will never be restarted.  If  
> you want that, shouldn't you use -S?  But if you really wanted a  
> migrated virtual machine to never be restarted, didn't you have a  
> problem before my patch as well?
>
> Note that autostart (the variable) is never read except to choose  
> whether to do that vm_start; so there can be no collateral effects from  
> passing -S.
>
> I don't understand what this patch is trying to achieve (and why).

If I pass --incoming to the VM, I don't expect the VM to start. I expect it
to wait, because I'll be handling it a migration later. However, it does start,
which is clearly wrong.

Of course, it does not start if we pass -S in conjunction with --incoming, but this
is a totally crazy semantics. An option should not depend on another option
to do the right thing.

As I said in the changelog, -S is still respected in migration because we're not
automatically vm_start()'in anything after migration completes, and also work
with a migrated machine that is already running. So what am I missing here?