A software defect that remained dormant for months, a routine network failover and a temporary reduction in server capacity collectively triggered a global outage.