VM failed to reboot after compute host reboot in Queens

[Jing Zhang]

This happens with VM with either tap (vif type ovs), or SRIOV (vif type hw_veb), or vhostuser (vif type vhostuser) ports in master (Queens) after nova/virt/libvirt/driver.py code change per https://git.openstack.org/cgit/openstack/nova/commit/?id=aaf37a26d6caa124f0cc6c3fe21bfdf58ccb8517 per https://bugs.launchpad.net/neutron/+bug/1744361.

This is because the above nova change only un-blocks VM from waiting network-vif-plugged event for vif type "bridge". The approach taken is questionable since it is error prone (has to unblock vif type one by one) and not future safe (new vif type can be added).

We propose either unblocks VM from waiting for network-if-plugged event for hard reboot completely, this is what we are doing now locally, or fix the root cause of missing network-if-plugged events in the host reboot scenario.

For comparison, we have checked Newton, we see exactly the same VM start up events in Newton, only network-vif-plugged event arrives as VM expects so VM is able to reboot successfully.

Please review and fix the issue.

[YAMAMOTO Takashi (yamamoto)]

the nova change in question has been reverted.
is it still an issue for some reasons, or are you suggesting a related improvements?
either ways, can you explain a little more?

[Jing Zhang (jing.zhang.nokia)]

After we move up to nova 17.03, we pick up the "revert" commit you mentioned above.
https://git.openstack.org/cgit/openstack/nova/commit/?id=2e03eae67d7977be2318dfb89be9890fa4ceb440

The issue is gone, it is good enough, no need for further churn.