Comment 3 for bug 1758316

After many logs rechecking, I think this is because the dvr and dvr_snat are both compute node:
https://github.com/openstack-infra/devstack-gate/blob/master/devstack-vm-gate.sh#L843

In the `Bug Description` LOG where the test vm was scheduled to the dvr_snat aka subnode-2:
http://logs.openstack.org/63/555263/3/check/neutron-tempest-plugin-dvr-multinode-scenario/e7c012f/logs/subnode-2/screen-n-cpu.txt.gz#_Mar_23_09_14_47_604613

That's why the tc rules were applied in "snat-namespace":
http://logs.openstack.org/63/555263/3/check/neutron-tempest-plugin-dvr-multinode-scenario/e7c012f/logs/subnode-2/screen-q-l3.txt.gz#_Mar_23_09_20_07_568567

DVR edge (or DVR edge ha) router will set the rules to snat-namespace:
https://github.com/openstack/neutron/blob/master/neutron/agent/l3/extensions/fip_qos.py#L223

I think, for dvr fip qos in such mixed deployment scenarios, we can install the rules again in the qrouter-namespce for dvr_snat node.