Open Stack

>> With tcpdump I can see instance making the ARP request 'who has 192.168.0.1'. However, the first answer is from bridge qvb, using the bridge mac address.
I believe this is due to L2 population + ARP responder mechanism where OVS itself creates an ARP reply and sends it back to avoid the ARP flooding. However, the reply should have the gateway’s MAC as its source. What source MAC do you see in the ARP reply?

Have you disabled reverse path filtering on your compute nodes (/etc/sysctl.conf) ?:
net.ipv4.conf.all.rp_filter=0
net.ipv4.conf.default.rp_filter=0


-Kaustubh
From: Jorge Luiz Correa [mailto:correajl at gmail.com]
Sent: Monday, May 16, 2016 3:35 PM
To: openstack at lists.openstack.org
Subject: [Openstack] Brigde on compute node confusing tenant network with system interface.

Hi list! I'm having problems with tenant network and one of compute nodes. I think that could be something with the compute node management ip address.

Context:

I'm testing the network scenario "classic with openvswitch" described here:

http://docs.openstack.org/liberty/networking-guide/scenario-classic-ovs.html

1 controller node, 1 network node and 2 compute nodes.

Compute node 1 has 4 interfaces:

p3p1 - 192.168.0.1 (management network)
p3p2 - 192.168.1.1 (tunnel network)
p4p2 - no address (vlan network)
em4 - no address (external network)

When I create a tenant network with network address 192.168.0.0/24<http://192.168.0.0/24>, instances launched on compute node 1 are having connectivity problem because they can't reach the network gateway 192.168.0.1.

Openvswitch agent creates a qbr bridge with a tap and a qvb interface. The instance has the address 192.168.0.5. When I try to ping 192.168.0.1 from this instance (its gateway), it doesn't work. With tcpdump I can see instance making the ARP request 'who has 192.168.0.1'. However, the first answer is from bridge qvb, using the bridge mac address. The correct would be the interface from virtual router, as I can see on compute node 2.

I think the IP address 192.168.0.1 on the management interface is making, for some reason, the bridge reply as if it was the gateway of tenant network.

This just occur with this address 192.168.0.1! If I create the tenant subnet with address 172.16.0.0/24<http://172.16.0.0/24> everything works great. Or, if I change the default setting for gateway address, make the gateway has another IP like 192.168.0.254, everything works fine too.

Can someone help me? Are there some configuration with sysctl to solve this? Somewhere to verify?

Tks!

- JLC
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20160516/f291fa67/attachment.html>