octavia/ovn: missed healthmon port cleanup
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
neutron |
Fix Released
|
Undecided
|
Fernando Royo |
Bug Description
Creating an octavia load-balancer with the ovn provider, adding a health-monitor and then members, octavia creates a neutron hm port in each subnet where a member was added.
Removing the members again, the hm ports do not get cleaned up. The hm removal then cleans up one of the hm ports, the one that is in the subnet where the vip happens to be. The others are still left and do not get cleaned up by octavia. This of course will cause issues when subnets can later not be deleted due to being still populated by the orphaned ports.
The cleanup logic simply does not match the hm port creation logic.
Mitigating factors:
* openstack loadbalancer delete --cascade does clean up all hm ports.
* Deleting the health mon before removing the members also avoids the issue.
tags: | added: ovn-octavia-provider |
Test script to reproduce the issue. /github. com/osism/ issues/ issues/ 921
Can be run against any OpenStack environment with octavia ovn provider loadbalancers without any special privileges.
But was originally observed against OpenStack 2023.2 (Bobcat) as configured by OSISM which uses kolla-ansible.
Original report is here:
https:/