Setup: three computers, ovs bridge on each of them with vxlan tunnels in full mesh topology. Send a broadcast packet to the internal port of the bridge.
Originally reproduced with openstack (see bug 1185521 comment 25 and following coments for details).
Happens at least with ovs 2.1.2 and 2.3.1.
The so far debugged symptoms are excessive flow dumping:
The leader revalidator thread is dumping flows from the kernel in a tight loop. The other revalidator threads and ovs-vswitchd thread are spending a lot of time on various locks (which is logical). The flow revalidation should not happen more often than every 500 msec but that's not the case here for some reason.
It seems the problem is seq_wait in udpif_revalidator which either sets poll_immediate_wake or causes poll_block to return immediately because of latch_wait(&waiter->thread->latch) in seq_wait__.
Strangely, I cannot reproduce it anymore. The storm itself is easily reproducible but the CPU usage I get is as expected.
Given we don't have anyone complaining, this has never been observed during normal operation and I can't reproduce anymore, I'm closing this.
If we hit this again, we'll reopen.