Some of the comments I made serve more to help my own memory jumping on and off the issue. ;) Basically the nodeinfo looks like libvirt ran into thinking that there is no node subtree in sysfs. Then it sets the number of nodes to 1 and directly scans the cpu subtree. I quickly did scan the upstream git tree of libvirt and did not find something obvious. I may have missed things or maybe it is the kernel that changed for the better. What I got in mind would be to make a special version of libvirt that has a lot logging around the area of obtaining the nodeinfo. If this can be tried on a non-production server, I think this will be less painful for all. If its ok for everyone to wait that long. I would go and prepare the test package(s) (Trusty/14.04 version) and post a link to them here. The debugging would be on error level, so the log level of libvirtd can be kept to errors (avoiding to fill up the log with other stuff).
-- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1446177 Title: Nodeinfo returns wrong NUMA topology / bad virtualization performance To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/libvirt/+bug/1446177/+subscriptions -- ubuntu-bugs mailing list ubuntu-bugs@lists.ubuntu.com https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs