OpenStack Compute (nova)

  1. Bug #1526818
  2. Comment #9

Comment 9 for bug 1526818

Revision history for this message
Michael Steffens (michael-steffens-b) wrote : Re: Incorrect and excess ARP responses in tenant subnets

Hi Augistina, hi Yushiro. I'm also sorry for the late reply! I had to wait for an opportunity to safely do the test, as our environment exposing the behavior got sort of "semi productive" ...

I executed

  sysctl -w net.ipv4.conf.all.arp_ignore=1

on all compute nodes and indeed that fixed the issue! Thanks a lot!

As far as Neutron is concerned, I believe that ML2 plugin and OVS plugin couldn't do much to fix the problem on their own? If so, I would suggest to add a note to

  http://docs.openstack.org/kilo/install-guide/install/apt/content/neutron-compute-node.html

to configure

  net.ipv4.conf.all.arp_ignore=1

in /etc/sysctl.conf, along with the other IP options.