← Back to team overview

yahoo-eng-team team mailing list archive

[Bug 1588393] Re: Switching OpenFlow interface to 'native' causes network loop

 

Reviewed:  https://review.openstack.org/325392
Committed: https://git.openstack.org/cgit/openstack/neutron/commit/?id=09ff5e5ebd2d608c7ac44ccab16d8e108d7181bc
Submitter: Jenkins
Branch:    master

commit 09ff5e5ebd2d608c7ac44ccab16d8e108d7181bc
Author: Ilya Chukhnakov <ichukhnakov@xxxxxxxxxxxx>
Date:   Fri Jun 3 18:57:15 2016 +0300

    Force "out-of-band" controller connection mode
    
    By default openvswitch uses "in-band" controller connection mode ([1])
    which adds hidden OpenFlow rules (only visible by issuing ovs-appctl
    bridge/dump-flows <br>) and leads to a network loop on br-tun when
    using native OpenFlow interface. As of now the OF controller is hosted
    locally with OVS which fits the "out-of-band" mode. If the remote OF
    controller is ever to be supported by openvswitch agent in the future,
    "In-Band Control" [1] should be taken into consideration for physical
    bridge only, but br-int and br-tun must be configured with the
    "out-of-band" controller connection mode.
    
    [1] https://github.com/openvswitch/ovs/blob/master/DESIGN.md
    
    Change-Id: I792a89d37b5d5319cc027835f6a1bfcbe7297ffb
    Closes-Bug: #1588393


** Changed in: neutron
       Status: In Progress => Fix Released

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to neutron.
https://bugs.launchpad.net/bugs/1588393

Title:
  Switching OpenFlow interface to 'native' causes network loop

Status in neutron:
  Fix Released

Bug description:
  * Description:
  After switching openvswitch agent to the 'native' OpenFlow interface (of_interface=native) the public network and the tunnel networks are flooded with ARP packets (see [1] for the tcpdump sample).

  * Environment:
   - DevStack stable/mitaka
   - 1 controller/compute and 2 compute nodes
   - configuration from [2]
   - ubuntu 14.04

  * How to reproduce:
  0. (WARNING) the following steps will flood the network, so it is recommended to use a virtual network as the provider network
  1. Deploy DevStack with access to the provider network (see [2]; 1 controller + 2 compute nodes)
  2. Set of_interface=native in the [ovs] section of /etc/neutron/plugins/ml2/ml2_conf.ini
  3. restart l2 agents on all nodes
  4. login to the default gateway and send a broadcast ARP request to the devstack's public network (arping -UD <gateway_ip>)

  * Expected result:
  normal network operation

  * Actual result:
  the public network and the tunnel network are flooded with ARP packets

  [1] http://paste.openstack.org/show/507292/
  [2] http://docs.openstack.org/developer/devstack/guides/neutron.html#devstack-configuration

To manage notifications about this bug go to:
https://bugs.launchpad.net/neutron/+bug/1588393/+subscriptions


References