ERROR: NetworkHostNotSet boot with --nic option

[Nachi Ueno]

ERROR: NetworkHostNotSet boot with --nic option

nova boot --image c11aec73-355b-4186-86fb-934bf6567e27 --flavor 4 --nic net-id=800e54c5-edf0-4e35-8b7e-271a724dacd5,v4-fixed-ip=192.168.21.3 test_instance

ERROR: NetworkHostNotSet: Host is not set to the network (9). (HTTP 400)

As Error message says, this is because Network host didn't set.
Same command will be success after instance creation without nic paramaters.

[Mark McLoughlin (markmc)]

Ok, I'm looking at NetworkManager.allocate_for_instance()

 networks = self._get_networks_for_instance(admin_context,
                                        instance_id, project_id,
                                        requested_networks=requested_networks)
        ...
        self._allocate_fixed_ips(admin_context, instance_id,
                                 host, networks, vpn=vpn,
                                 requested_networks=requested_networks)

get_networks_for_instance() looks up the networks using network_get_all_by_uuids() and this is what throws NetworkHostNotSet if the network isn't associated with a host

However, it looks to me like the network will only be associated with a host by the set_network_host() in RPCAllocateFixedIP._allocate_fixed_ips()

i.e. we're requiring the network to be associated with a host before actually doing the association

Marking Confirmed, something looks wrong here

Nachi - could you provide some more details on exactly what version you're using, whether you can still reproduce, what network manager you're using, whether you have enabled multi_host etc.? Also the exact steps to reproduce - e.g. I think this would only happen if the user requests a specific network when that network was never used before

[Yovko Yovkov (yyovkov)]

Hi Mark,

I'm facing the same issue on CentOS 6 and the packages are:
openstack-dashboard-2012.1-4.el6.noarch
openstack-utils-2012.1-2.el6.noarch
openstack-keystone-2012.1-8.el6.noarch
openstack-glance-2012.1.1-1.el6.noarch
openstack-nova-2012.1.1-3.el6.noarch

Regards,
Yovko Yovkov

[melanie witt (melwitt)]

@markmc,

I have seen this happen in Essex using FlatManager and networks created with --multi_host=T.

I compared the code you mentioned:

Essex: https://github.com/openstack/nova/blob/stable/essex/nova/db/sqlalchemy/api.py#L1951

Folsom: https://github.com/openstack/nova/blob/stable/folsom/nova/db/sqlalchemy/api.py#L2164

so it seems this shouldn't be a problem anymore in Folsom.

[Vish Ishaya (vishvananda)]

I believe the code from this patch will fix it:

https://github.com/openstack/nova/commit/7937144fce54570b2da543663e6ee5e64b1c3cdb

It might be a little tricky to backport it to essex and is probably out of scope for stable/essex.

[Vish Ishaya (vishvananda)]

Fixed by: https://github.com/openstack/nova/commit/7937144fce54570b2da543663e6ee5e64b1c3cdb

[Vish Ishaya (vishvananda)]

Fixed by: https://review.openstack.org/12749