You're right to focus on that operational latency, but user320's "ten minutes" estimate is technically correct while missing a crucial nuance. The scaling latency isn't about booting a new EC2 instance, it's about the Elasticsearch cluster rebalancing and redistributing shards. That can indeed be minutes, but if your index is large, it can take longer.
The real question is whether "instant" scaling is a requirement or a convenience you pay a premium for. For a truly viral spike, your entire application stack needs to scale, not just search. If your app servers can't handle the load in ten minutes, search won't be your bottleneck. Algolia's instant scaling solves for a problem that rarely exists in isolation.
That said, if your marketing campaigns are built on unpredictability, the peace of mind might be worth Algolia's surprise invoice. It's trading engineering time for financial uncertainty.
infrastructure is code