qemu-devel
[Top][All Lists]
Advanced

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

Re: [Qemu-devel] [PATCH 2/10] Allow the monitor to be suspended during n


From: Anthony Liguori
Subject: Re: [Qemu-devel] [PATCH 2/10] Allow the monitor to be suspended during non-blocking op
Date: Wed, 10 Sep 2008 10:40:54 -0500
User-agent: Thunderbird 2.0.0.16 (X11/20080723)

Avi Kivity wrote:
Daniel P. Berrange wrote:

Or rather, such state should be part of the migration. There's the question of whether to transform the path on the target, but "which media is in the drive" is part of the hardware state.

(logically we would copy all of the data of all block devices, but that's not very practical, so we assume shared storage).

What other commands are unsafe during migration? I exclude host device assignment which is obviously migration unfriendly.

USB + virtio device attach/detach - well at least have the same need as
media change - you'd need to propagate the change to the other side in
some way. Currently migrate just assumes the management tool/admin has
started QEMU on the destinations with the matching hardware config, and
for libvirt to use the monitor to add USB /virtio devices at both ends
has the race condition/synchronization problem .

Any hotplug, for that matter.

The hardware topology should be part of the state; there is no other way to deal with hotplug. Hotplug can create configurations that are impossible to recreate using the command line.

Yes, I agree. I'd like to introduce a "machine" section that contains nothing but a version id. v1 will simply be a placeholder. If we add new hardware to the machine type, we'll bump to v2. When we merge hot plug support, we need to add actual contents to the save section describing any hot plugged hardware.

Regards,

Anthony Liguori

In general a hardware device has two parts: a guest visible part and a host interface part. For networking, that's easily visible (-net nic and -net tap/user). For block devices, the filename and caching mode is the host interface while, the interface type and index is the guest part.

Migration should transfer the guest part, and the host parts should be specified from the command line or monitor on the migration target, as they can change from host to host.






reply via email to

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