Comment 7 for bug 1524898

We now have a job which has failed with the custom QEMU present

http://logs.openstack.org/03/335903/1/check/gate-tempest-dsvm-multinode-live-migration/5bae103/

See the source QEMU log:

http://logs.openstack.org/03/335903/1/check/gate-tempest-dsvm-multinode-live-migration/5bae103/logs/subnode-2/libvirt/qemu/instance-00000003.txt.gz

migration_completion: global_state_store ret=0
migration_completion: vm_stop_force_state ret=-5

This message with ret=-5 is reporting the return value of the 'vm_stop' API call, which in turn is reporting do_vm_stop which in turn is reporting the value from bdrv_flush_all(). The latter reports errno values, and -5 corresponds to EIO.

So, we're failing the migration switch over because the block driver was unable to flush outstanding I/O for some unknown reason.