More than 50% of large distributed system downtime is caused by configuration errors, failed software updates, or network attacks.

With configuration error, or failed software update everything can go down at once, or there are cascades of restarts that kill the system.