Agent API gets broken if OVN DB is upgraded to the one introducing chassis_private

[Jakub Libosvar]

Neutron determines which table to use for agent healthchecks at the initialization phase. With OVN databases that don't implement chassis_private table a chassis table is used, which has scaling impacts.

However, if OVN DB is upgraded while Neutron is running, the healthchecks will be still performed on chassis table while OVN already uses chassis_private. This leads to reporting dead agents by the API.

We should detect OVN was upgraded and start using chassis_private table since.

[Bence Romsics (bence-romsics)]

Is the use of chassis_private always better than using the chassis table? If it is then alternatively we could just upgrade this schema on the next neutron start unconditionally. That may be simpler than detecting the upgrade and all schemas would be uniform after. Just floating the idea...

[Jakub Libosvar (libosvar)]

Yes, chassis_private improves scalability big time. The schema is defined in OVN Db processes, the problem is only when those processes are restarted on new schema while neutron-server keeps running. Restarting neutron-server fixes the problem.

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: master
Review: https://review.opendev.org/760962

[OpenStack Infra (hudson-openstack)]

Related fix proposed to branch: master
Review: https://review.opendev.org/760963

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.opendev.org/760967

[OpenStack Infra (hudson-openstack)]

Change abandoned by Jakub Libosvar (<email address hidden>) on branch: master
Review: https://review.opendev.org/760963

[OpenStack Infra (hudson-openstack)]

Change abandoned by Jakub Libosvar (<email address hidden>) on branch: master
Review: https://review.opendev.org/760962

[Lucas Alvares Gomes (lucasagomes)]

I think this has been fixed by https://review.opendev.org/c/openstack/neutron/+/752795/ ?

[Jakub Libosvar (libosvar)]

Yes, I think it's https://review.opendev.org/c/openstack/neutron/+/760967 but you're right, it's fixed now. Thanks for pointing it out!

[OpenStack Infra (hudson-openstack)]

This issue was fixed in the openstack/neutron 18.0.0.0rc1 release candidate.