Similar to, but slightly different from bug 1712444, we have found the upgrade of the openvswitch-switch package by unattended-upgrade (or otherwise) will trigger the service restart of openvswitch-switch within the neutron-openvswitch charm if the config-changed hook is called.
While this is a reasonable behavior based on an assumption that if the /etc/default/openvswitch-switch file changes due to upgrade and the charm resets it to the charm-configured version of the file, we should want to restart the service to be on the latest code. However, the restart of the service causes between 6-12 seconds of network outage for the tenant VMs utilizing OVS.
Would it be possible to have a config-flag to disable the charm's ability to restart the openvswitch-switch service outside of the install hook to avoid automated network outages due to package upgrades?
Elsewise, is there a way to serialize the resulting restarts in such a way that only one member of the neutron-openvswitch application/service is restarting at a time along with a buffer to allow for high availability applications to failover and fail back and not be afflicted by multiple nodes' switches being restarted simultaneously.
Similar to, but slightly different from bug 1712444, we have found the upgrade of the openvswitch-switch package by unattended-upgrade (or otherwise) will trigger the service restart of openvswitch-switch within the neutron-openvswitch charm if the config-changed hook is called.
While this is a reasonable behavior based on an assumption that if the /etc/default/ openvswitch- switch file changes due to upgrade and the charm resets it to the charm-configured version of the file, we should want to restart the service to be on the latest code. However, the restart of the service causes between 6-12 seconds of network outage for the tenant VMs utilizing OVS.
Would it be possible to have a config-flag to disable the charm's ability to restart the openvswitch-switch service outside of the install hook to avoid automated network outages due to package upgrades?
Elsewise, is there a way to serialize the resulting restarts in such a way that only one member of the neutron-openvswitch application/service is restarting at a time along with a buffer to allow for high availability applications to failover and fail back and not be afflicted by multiple nodes' switches being restarted simultaneously.