On 04/17/2012 02:32 AM, Jiri Denemark wrote:
In the meantime, I think I will patch this instead by documentation: state that non-zero bandwidth in virDomainBlockRebase() may be silently ignored; if error checking is needed, the user should use a separate call to virDomainBlockJobSetSpeed(); and to see if a request was honored, use virDomainGetBlockJobInfo() (with the known race that the job might already be complete).
Makes sense, although we don't want to store the result of BLOCK_JOB_SPEED_INTERNAL to be stored in ret, then.
Actually, qemu may save us - they are agreeing that this is a problem, and will probably have a patch before libvirt 0.9.12 that allows block-job-set-speed to be called prior to block-stream or drive-mirror; at which point the correct fix is to then try to set the speed _prior_ to starting the job (but only when we know we are talking to qemu 1.1, not RHEL 6.2). It would leave RHEL 6.2 with the race, but oh well. -- Eric Blake eblake@redhat.com +1-919-301-3266 Libvirt virtualization library http://libvirt.org