On 08/04/2013 09:57 AM, Perry Myers wrote:
Hi,
I followed the instructions at:
http://openstack.redhat.com/Neutron-Quickstart
http://openstack.redhat.com/Running_an_instance_with_Neutron
I ran this on a RHEL 6.4 VM with latest updates from 6.4.z. I made sure
to install the netns enabled kernel from RDO repos and reboot with that
kernel before running packstack so that I didn't need to reboot the VM
after the packstack install (and have br-ex disappear)
The packstack install went without incident. And I was able to follow
the launch an instance instructions.
Ok, retried this but took advice from folks on this thread.
Since l3 agent and dhcp agent in RDO are not right (they comment out
ovs_use_veth=True and veths are required for the netns support in RHEL
kernels)
marun summarized this nicely:
"if ovs_use_veth is set to false, a regular interface and an internal
ovs port will be used, and the regular interface will be moved to a
namespace during setup. if ovs_use_veth is set to true, a veth pair will
be used with one endpoint created in the namespace. it is a limitation
of rhel's netns implementation that requires the second approach, as
virtual interfaces can only be created in namespaces, not moved
post-creation."
With manually enabling ovs_use_veth=True for l3 and dhcp agents, I was
able to get cirros VM to get an ip address on launching.
What doesn't work now is pinging/sshing to the floating ip address from
the host (which is itself a VM)
Yes, I did open those ports in the default security group, and I also
made sure the instance was launched with the default security group.
But that being said, I wanted to check the logs to see if some of the
previous errors went away. dhcp-agent and l3 agent logs look clean now
(aside from the amqp initial connection errors)
My next test will be to run this exact same scenario but with
NetworkManager disabled.
Perry