TreasureData deals with huge amount of streaming data import request and saves them into our database in realtime without any lost.

New scalable system is required to process requests increasing day by day, and I decided to replace old system with rails and fluentd to new system with h2o and mruby for gateway server.

I'll introduce why h2o and mruby is good and how I optimized mruby server handler for the system. In addition, I'll talk about my patches for h2o to make it possible for mruby parallel processing/asynchronous processing. And also, I'll show benchmarks of actual product.