On Thu, Oct 08, 2015 at 12:08:58 +0100, Daniel P. Berrange wrote:
On Thu, Oct 08, 2015 at 12:06:09PM +0100, Richard W.M. Jones wrote:
>
> I wonder if there's a reason why it is not possible to dump the core
> of a KVM domain which is set to autodestroy.
>
> $ virsh dump guestfs-4xe592eyqq4rfnk9 /tmp/guestfs-4xe592eyqq4rfnk9.core
> error: Failed to core dump domain guestfs-4xe592eyqq4rfnk9 to
/tmp/guestfs-4xe592eyqq4rfnk9.core
> error: Requested operation is not valid: domain is marked for auto destroy
>
> All libguestfs domains have the autodestroy flag, but I am trying to
> debug one which is crashing in the guest (RHBZ#1269779) and having the
> ability to dump it would be very helpful.
>
> I added the following patch to libvirt, which allowed me to dump the
> domain with no apparent ill effects.
Yep, I see no reason to block use of the core dump API. Looking at the
patch I think this is probably just an over-sight - ie it is merely
blocked as an accidental side-effect of us blocking migration.
I think you should actually use virsh dump --memory-only ... to go
through dump-guest-memory instead of migration. This should create an
elf formated core which crash supports better and even on ppc64, for
example.
Jirka