Hi, I'm part of a team running a small:ish cloud - we have an issue atm where neutron thinks that some hosts exists and should be linked. This results in a arp flood - which on it's own is not a problem.
As soon as we get "neighbour: arp_cache: neighbor table overflow!" the machine stops responding to connections.. Even if this feels like a bit of a special case, it also feels *very* wrong ;) It's like the "I can't find" queries add to the cache without adding a entry - arp -a - n |wc -l is only 149 f.ex. Anyone with ideas or clues?