On 02/27/2013 07:14 PM, Wenchao Xia wrote: > δΊ 2013-2-28 0:08, Markus Armbruster ει: >> Wenchao Xia <xiaw...@linux.vnet.ibm.com> writes: >> >>> This interface now return valid internal snapshots for whole vm. >>> Note that filter use bdrv_can_read_snapshot() instead of >>> bdrv_can_snapshot(), which should be the correct behavior in information >>> retrieving funtion. >> >> This is just like query-images, except it's for the system as a whole >> rather than individual block devices. Correct? >> > Yes, just for snapshot info for whole vm.
Another way of looking at it: if a guest has two images, a.qcow and b.qcow, where a.qcow includes snapshot id '1' named 'bar', and id '2' named 'baz'; and where b.qcow includes snapshot id '1' named 'foo', and id '2' named 'bar', then querying individual block devices will see both internal snapshots of each device, but querying the system as a whole will see only a single snapshot named 'bar' and with no reliable id. This is because the 'loadvm' HMP command (or whatever equivalent we finally ever make in QMP) needs to act on all devices at once; reverting to id '1' or name 'foo' doesn't make much sense, because those are not consistent. A management app might be able to piece things together through a more complicated disk-at-a-time transaction, at a lower level than what 'loadvm' does. But this command seems useful for stating what snapshots are consistent enough for 'loadvm' to have a chance of succeeding. -- Eric Blake eblake redhat com +1-919-301-3266 Libvirt virtualization library http://libvirt.org
signature.asc
Description: OpenPGP digital signature