In the talk, I will present several common traps that regular backend engineers are falling when integrating with Elasticsearch and measures that could prevent them. First, I'm going to present what was the situation when throwing more metal into the Elasticsearch performance problem stopped helping. Second, I'll present several optimizations that actually made the situation better and lessons learned from it. Third, I'll present the current situation of Elasticsearch deployment. The talk will include benchmarks with real-world numbers collected at Vinted. An example of a lesson learned: Ruby on Rails by default stores IDs in the database as auto-increment integers, while Elasticsearch is much faster when doing `terms` query on `keywords`. By how much faster? Come to the presentation.