My company provides dynamic content in emails, and as such gets large traffic spikes when 10 million emails get sent at once and everyone begins opening them.

My company provides dynamic content in emails, and as such gets large traffic spikes when 10 million emails get sent at once and everyone begins opening them. The content's configuration (in postgres) is trivially cacheable, but our app servers render different content based on the user's context. So we have a bunch of shared-nothing app servers that we can spin up and down based on the emails we know are going out. Automatically detecting spikes and spinning up new instances between the send and the peak is much harder, though.