In my consideration , all services should not affect others while restart/stop service within cloud platform...
That means instances connection should be alive while admin restart nova-network host , but it won't work as your expectation. I faced this issue one month ago. And I spend around 1 hour to understand what's going on with this problem.
In regular nova deployment , nova-network host is the gateway of all instances. This Linux network box works like a router. So that has a ARP table over the box.
In my test , if the flat_interface(or vlan_interface) do not auto up after reboot , the box will lose ARP table. And you can not ping or ssh instance anymore. While you up the nic manually , you have to wait for ARP rebuild .
There's a correct ARP info
Address HWtype HWaddress Flags Mask Iface 192.168.1.1 ether 00:13:49:d0:dd:9c C eth0
Sometimes , It's really crazy for me . I ever wait for over 24 hours.
ARPING will help you discover the host.
If the network configuration is correct after reboot . It never happened to me.
In the above environment , I always up the flat_interface manually . Luckily , the ARP table is complete with auto up the flat_interface.
Due to the flat_interface which is eth1 need to be auto-up without any network parameters for nova-network to handle the bridge configuration. There's the approach to auto-up nic at boot up.
Add to /etc/network/interfaces
auto eth1 iface inet eth1 static address 0.0.0.0 netmask 0.0.0.0
Another code less option(From Stackops )
up ifconfig eth1 0.0.0.0
Advanced skill(From Mr. Unknown)
add this in the pre-start script:
ip link set eth1 up
As the kindly mention:
This would be rather better, nova-network will start only when eth1 has been activated.
And remember, the compute nodes also need to create bridge,you can add the same command in /etc/init/nova-compute.conf on compute nodes.