libvirt: live-migration crash with segfault while transfering memstate

[Alex Walender]

Description
===========

Using recent Ubuntu Stein Cloud Packages, we are observing random live-migration crashes on the target host. Libvirt is having a SEGFAULT on the qemu driver. Transferring block devices usually works without issues. However, the following memory transfer is causing the target libvirtd randomly to close down its socket, resulting in a roll-backed migration process.

Libvirt log on target host before the crash is attached.

Steps to reproduce
==================

- Start a live-migration with block mode between 2 identical hosts.
- Wait until transfer of blockdisks is done
- During memory transfer, target host libvirt crashes.

Expected result
===============

Live-Migration completes onto the new host as intended.

Actual result
=============

Target host libvirtd crashes with SEGFAULT, causing a rollback of the migration.

Environment
===========

Ubuntu 18.04.4 LTS (GNU/Linux 4.15.0-99-generic x86_64)
OpenStack Stein (Ubuntu Cloud Archive)
Libvirt+QEMU_x86

keystone-common 2:15.0.1-0ubuntu1~cloud0
libvirt-daemon 5.0.0-1ubuntu2.6~cloud0
qemu-system-x86 1:3.1+dfsg-2ubuntu3.7~cloud0
neutron-linuxbridge-agent 2:14.2.0-0ubuntu1~cloud0
neutron-plugin-ml2 2:14.2.0-0ubuntu1~cloud0
nova-compute 2:19.2.0-0ubuntu1~cloud0
nova-compute-libvirt 2:19.2.0-0ubuntu1~cloud0
python-rbd 14.2.10-1bionic
python3-cinderclient 1:4.1.0-0ubuntu1~cloud0
python3-designateclient 2.9.0-0ubuntu1
python3-glanceclient 1:2.16.0-0ubuntu1~cloud0
python3-neutronclient 1:6.11.0-0ubuntu1~cloud0
python3-novaclient 2:13.0.0-0ubuntu1~cloud0

[Alex Walender (awalende)]

Captured this in debug logs

[Balazs Gibizer (balazs-gibizer)]

Based on the logs I don't think this is a nova bug. I suggest to contact the libvirt developers at https://libvirt.org/bugs.html

I'm setting this as Invalid if further investigation on the libvirt side indicates that somehow nova causes the libvirt segfault then feel free to set this bugreport back to New.