On 06/08/2013 12:48 AM, Kevin Wolf wrote: > Hm, yes, I was confused. :-) > > Anyway, the options stay the same: Either C1 must somehow inherit the VM > state from A on the Sheepdog level, or we must make sure to get this > order: > > 1. Switch to (read-only) snapshot A > 2. Load the VM state > 3. Create (writable) snapshot C1 and switch to it > > This is a different order as required for qcow2 (where C1 would inherit > the VM state from A, so our current first 1, then 3, then 2 works), but > if we can't fix it on the Sheepdog side, we'll have to touch the core > code of qemu. I don't like much to touch the block.c code for it, but if > it's required to fix a bug, let's just do it.
Hmm, I found actually we can inherit A's vdi_id from C1. I've sent v3 and I think this version will fix the possible bug you pointed out and do loadvm the way core code assumes. Thanks, Yuan
