When user create network with isolated subnet, dhcp may create metadata ns proxy with router id.
How to reproduce:
1. create a router: R1
2. create a network: Net1
3. create two subnetworks: Sub1, Sub2
4. attach Sub1 to R1. (do not attach Sub2)
if you deploy dhcp-agent and l3-agent on seperate node, you will see:
a) dhcp-agent will start metadata ns proxy with router-uuid with port 80 (This metadata ns proxy will not work in data path)
Here is a sample, you can see the metadata_port is 80 with router_id:
neutron 237535 0.0 0.2 295864 46284 ? S Jul07 0:00 /usr/bin/python2 /bin/neutron-ns-metadata-proxy --pid_file=/var/lib/neutron/external/pids/160f2356-3bf5-4a1c-80e6-b9ef8b971047.pid --metadata_proxy_socket=/var/lib/neutron/metadata_proxy --router_id=160f2356-3bf5-4a1c-80e6-b9ef8b971047 --state_path=/var/lib/neutron --metadata_port=80 --metadata_proxy_user=995 --metadata_proxy_group=992 --debug --log-file=neutron-ns-metadata-proxy-160f2356-3bf5-4a1c-80e6-b9ef8b971047.log --log-dir=/var/log/neutron
b) l3-agent will start metadata ns proxy with router-uuid with port 9697 (This metadata ns proxy works in data path)
if you deploy deploy dhcp-agent and l3-agent on same node, metadata ns proxy may start by dhcp or l3 agent, both process manager in dhcp agent and l3 agent will monitor this metadata ns proxy. This means if you kill this metadata ns proxy, it may be start up by dhcp agent or l3 agent, this is relay on which agent start it first. you will see the port of metadata ns proxy may be 80 or 9697. The action is unpredictable.
The problem is DHCP agent should not manage metadata ns proxy with router_id, this kind of metadata ns proxy should be managed by l3 agent.
When dhcp agent find isolated subnet, it should start metadata ns proxy with network_id, not router_id, the following code should be removed:
https://github.com/openstack/neutron/blob/master/neutron/agent/dhcp/agent.py#L508
Fix proposed to branch: master
Review: https:/