I was setting up a VM independent of autopkgtests to debug this further, but so far only confirmed what I had already: Yet as documentation if we later on need to refer to that data:
Logs of journalctl -fex and the libvirt log on sl guest when doing the killing restart # Before: $ virsh list --all Id Name State ---------------------------------------------------- 3807 sl running # Then on restart $ /etc/init.d/libvirtd restart # /var/log/libvirt/lxc/sl.log: 3807: info : hostname: zesty-test-smoke-lxc 3807: error : virNetSocketReadWire:1801 : End of file while reading data: Input/output error # journalctl -fex: [...] normal stopping / starting message - then after "libvirtd.service has finished starting up": libvirtd[4073]: internal error: No valid cgroup for machine sl audit[4125]: AVC apparmor="STATUS" operation="profile_remove" profile="unconfined" name="libvirt-354e2b39-9c30-4080-9323-23d27f8d2cf8" pid=4125 comm="apparmor_parser" kernel: audit: type=1400 audit(1482226905.416:21): apparmor="STATUS" operation="profile_remove" profile="unconfined" name="libvirt-354e2b39-9c30-4080-9323-23d27f8d2cf8" pid=4125 comm="apparmor_parser" libvirtd[4073]: End of file while reading data: Input/output error # The guest appears as shut off now (sometimes I have to restart twice to get the guest listed here again): $ virsh list --all Id Name State ---------------------------------------------------- - sl shut off So the "root cause" still appears as: "internal error: No valid cgroup for machine sl" After that it is on the kill path, removing the profile but no other error shows up. I saved some /etc/libvirt and cgroup info. But no content in any of those changed between bad case (after first restart) and good case (after staring the guest once again). I don't see where it is "fixing itself up" after starting the guest again :-/ And I'd have hoped that this would be my entry point to find a fix.
