Investigate increasing the default ovn-controller and SBDB probe intervals. The default is currently 5 seconds for both, but Russell says every single non-trivial deployment hits the 5 second limit and drops connections continuously, requiring eg OpenStack to default to 180 seconds. Perhaps our defaults in OVN upstream/downstream should be changed. Or dynamically change the probe interval in ovn-controller based on an average of the last recompute times + some amount of slack.
Dynamically changing the probe interval seems a very good idea.