How about network partitions across availability zones? Happens all the time for us, so much in fact that we had to build a tool to test connectivity across AZs just to correlate outages and find the smoking gun.