Firebolt: Asynchronous Processing System for Payments
Submitted by Barani Subramanian (@baranis30) on Thursday, 21 February 2019
Section: Crisp talk Technical level: Intermediate
At Flipkart Payments we build robust systems to provide a smooth experience to customers. In this session, we’ll talk about how we levarage aysnchronous processing to improve the performance of the payment systems which are pivotal to Flipkart. We’ll go over the details of the streaming pipelines we built using MySQL, Hbase, Kafka and Storm, and the customization we did over them. We’ll share our journey and learnings for the same.
Firebolt: Async Processing for Payments - System functionlity and components - High level architecture Why Async Processing - Performance impact of sync processing in user live path - Customizations for high throughput Resilient Streaming - Avoiding Head of Line blocking through sidelines - Automatic sideline management Reliable Pipelines - Transactional event processing with high availability - Achieving at-least once in-order delivery for database relayers - Replay capability with GTID - Monitoring the system
Barani Subramanian works as a software developer in Flipkart. She has 4+ years of experience in Sales and Payments domains. She likes to read books in free time.