Description of problem:
When spice migration is initiated, rhevm instructs libvirt on migration source machine (via vdsm) to migrate the VM to the destination host, specifying among others libvirt uri of destination. Source libvirt then connects to the destination libvirt that will then tell the source connection details. The source libvirt in turn instructs connected client (via qemu) connection details such as hostname, port and others.
This configuration works in stock RHEV setups but it will break in more complex ones:
* if hostname of destination host does not resolve to the address of
display network interface (IOW if display network is configured to
be separate from management network)
* if upcoming display network address override is used (bug 788977)
The correct solution to all these woes is for RHEV-M to get display network fqdn/ip out of its database and pass it via vdsm to the source libvirt, that will in turn instruct client to connect to the right host.
Version-Release number of selected component (if applicable):
3.0-3.2 (all version that use libvirt at least)
How reproducible:
always
Steps to Reproduce:
1. set up display network separate to the management network, run VM
2. connect to the VM with a client
3. migrate the VM
Actual results:
client gets disconnected
Expected results:
client stays connected
Additional info: