Additional fixed ip address assigned to vm after nova-network restart
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Compute (nova) |
Won't Fix
|
Low
|
Unassigned |
Bug Description
So I'm runing into a problem when I restart nova-network on a compute node in a multi_host setup. Each time I
restart nova-network on one of the compute hosts, multiple guests on the other host are allocated additional
fixed ip addresses in addition to the one they already have and a stack trace is thrown in network.log. The
nova-br100.conf is updated to include both addresses on the same mac/hostname. I'm still able to access the
vm just fine on thier original ip address.
Recreating the problem is very consistent:
1) From a fresh install, bring up 3 virtual machines and take down the first one that came up (typically
instance_id 1 in the networks table).
2) Stop nova-network on either of the compute hosts. This will produce a stack trace in the network.log on
the other compute node complaining about 'InstanceNotFound: Instance 1 could not be found'
3) Look at horizon, dnsmasq's config (nova-br100.conf) or the networks table in the database and you will
see an additional ip address allocated to the vm that was running on the host that did not have nova-network
taken down. The ec2 compatibility api only reports the new ip address for the instance.
My setup:
I'm running 2 hosts for testing, one is the compute host running nova-api, nova-compute and nova-network. The
other is the controller running nova-api, nova-compute, nova-network as well as the other needed services
(keystone, glance, scheduler). I'm running a fixed network only (192.168.97.0/24), no floating, in a multi_host
setup with the routing handled by an external router (setup with dhcp flags for dnsmasq) with eth1 being the vm
network. Basically nova-network and nova-api are there only for dhcp and meta-data.
I'm using essex packages from EPEL on Centos 6.2. On the compute nodes:
python-
python-
openstack-
dhcp-options.conf (flags for dnsmasq):
dhcp-option-
#dhcp-option-
dhcp-option-
nova-br100.conf after the problem:
fa:16:3e:
fa:16:3e:
View of the networks table before the problem:
http://
View of the networks table after the problem:
http://
My nova.conf (it's the same on both hosts except for my_ip and the console addresses):
http://
I'm wondering if this happens because you set force_dhcp_ release= False.
I think setting it to True will make what you describe here go away.