In the world of modern cloud computing, horizontal scaling normally stands for adding/removing instances are per the current application load. However, normally launching of a new instance normally takes one or two minutes and in case of extremely sudden load increase (or even a DOS attack) such scaling may not be fast enough and lead to existing application processes dying. How would you mitigate such cases ? Would you keep a few spare nodes ? Would you predict the increase (if possible) ? Or prepare for possible events (such as Black Friday's) by adding additional machine power in advance ?
If possible, I would like to hear on possible solutions, as well as the comments on the questions listed above.
This question might be too broad, and I will edit/elaborate if required.