SNAT might be lost on a subnet when deleting a VM

[Oleg Bondarev]

When a VM is deleted and if it's the last VM from a particular subnet on this compute host - neutron server will delete DVR router from that host.
Due to a bug in some cases this may lead to SNAT portion deletion for this subnet (external access is lost for VMs, snat port is in DOWN state) - this is when SNAT is scheduled on a controller that has no DHCP ports bound from the subnet.

The bug is a result of combination of https://review.fuel-infra.org/#/c/10576 and https://review.fuel-infra.org/#/c/20143 where the former has some customization comparing to upstream (evil!).

[Oleg Bondarev (obondarev)]

Following patch should fix the issue (as a side effect mostly): https://review.openstack.org/#/c/252855 - need to backport

[Fuel Devops McRobotson (fuel-devops-robot)]

Fix proposed to branch: openstack-ci/fuel-7.0/2015.1.0
Change author: Oleg Bondarev <email address hidden>
Review: https://review.fuel-infra.org/23829

[Denis Meltsaykin (dmeltsaykin)]

Marking as Invalid for 8.0, as the fix is merged in stable/liberty.

[Oleg Bondarev (obondarev)]

Only valid for MOS 7

[Ekaterina Shutova (eshutova)]

Verified on MOS 7.0 + mu5 updates.

Configuration: 3controllers + 2 computes, DVR.
Network and distributed router created in such way that dhcp was on 2 controllers and snat was on third one.
Created 2 instances on different computes. Checked traffic. After deletion of one instance, SNAT wasn't lost, it stayed on the same controller as before and pings were sent to the external network from not-deleted VM.