openstack team mailing list archive

[Maciej Gałkiewicz <maciejgalkiewicz@xxxxxxxxxxxxx>]

Hi

Thx for such detailed response. I have approach at least three problems
with the migration.

1. Controller node (nova-api) is probably not able to speak with
nova-network and quantum simultaneously. Old instances loss their IPs (at
least in dashboard) which forces me to migrate all instances at once.

2. I am using gre tunneling and openvswitch with per instance private
networks. Each tenant will have several networks, each instance will use
only one network. In my opinion virtual link between instance (eth0) and
compute node (tap interface bridged with br-int) should be in generated
VLAN. In my configuration quantum ovs plugin creates tap interface in
trunking mode. DHCP server on network node is only listening on interface
assigned to specified VLAN. Summing up DHCP is properly configured for for
example VLAN 5 but instances which should have been in VLAN 5 are in native
VLAN due to trunking mode on tap interface on compute node.

3. I would like to connect instances to Internet through double NAT. I dont
like the idea of bridging eth0 on network node with br-ex because I have
only one routeable IP address assinged to this node. The problem is that
while doing double NAT something goes wrong and only first IP packet is
natted and the rest is not (simple ping to outside world). I wasnt able to
debug it better with iptables, conntrack, routing table cache.

regards
Maciej Gałkiewicz