Nova compute container broken after upgrade

[Mark Goddard]

After performing an upgrade of the nova_compute service via 'kolla-ansible upgrade', the container can end up in a strange broken state, where the service is not functioning.

I believe this is caused by the SIGHUP [1] sent to each of the nova containers. A recent change [2] proposed to kolla-ansible it was suggested that oslo.service treats SIGHUP as SIGTERM for foreground processes, which could be the cause of the issue. In kolla bug 1799642, it is stated that signals sent to the container are automatically forwarded to all child processes, which could also be an issue.

The code should be fixed to ensure that the service handles SIGHUP as expected, or is restarted cleanly.

[1] https://github.com/openstack/kolla-ansible/blob/master/ansible/roles/nova/tasks/reload.yml
[2] https://review.openstack.org/#/c/614468/1

[Mark Goddard (mgoddard)]

Fix merged to master, will land in Stein: https://review.openstack.org/645614

[Mark Goddard (mgoddard)]

Backports:

Rocky: https://review.openstack.org/647700
Queens: https://review.openstack.org/647701
Pike: https://review.openstack.org/647702