While attempting a live migration tunnelled like this :
virsh migrate --live --tunnelled --p2p guest1 qemu+ssh://192.168.99.3/system
the source host (node1) logs this error :
Oct 6 17:44:06 node2 libvirtd: 17:44:06.513: error : qemuMonitorTextGetMigrationStatus:982 : internal error cannot parse migration data transferred statistic 1052 kbytes#015#012remaining ram: 539972 kbytes#015#012total ram: 541024 kbytes#015#012
And the guest VM freezes and start consuming 100% CPU infinitely.
=========================================================
SRU Justification:
1. Impact: tunnelled migration fails
2. Development fix: consists of a patch which fixes the parsing of the 'info migration' reply.
3. Stable Fix: a simple cherrypick of the development fix
4. Test case:
1. Set up two hosts with libvirt-bin.
2. Work around bug 869553. You can do this by simply turning off apparmor on both.
3. Have them share storage (I install nfs-kernel-server on one, nfs-client on the other, place /mnt on the server's /etc/exports, and mount it on the client)
4. Create a VM with disks on the shared storage. I basically follow https://wiki.ubuntu.com/SergeHallyn_libvirtnest but replace the disk and cdrom images under /mnt.
5. Create a ssh key for root on the client; place it in root's .ssh/authorized_keys on the server. Test that root on the client can ssh to the server. Restart libvirt-bin
6. start the VM on the client
7. Migrate using:
virsh migrate --live --tunnelled --p2p cdboot qemu+ssh://server/system
8. With the patch, the vm should now be running on the server and not the client.
5. Regression potential: this patch only affects the migration code, so if there is a regression it should only affect users of (non-tunnelled, as tunnelled is broken anyway) migration.
=========================================================
When looking at the upstream git I found that Daniel P. Berrange had already fixed that in commit 0d3eee7fe8bcaa49. I have extract only the portion of this commit that touches src/qemu/