Eventual Consistency ftw.
summary
Klarna migrates a monolith into a no-master two-site architecture that accepts purchases even when back-end processing is offline, trading strict synchronization for uptime; the company already serves over 15,000 e‑commerce sites with >100M EUR turnover and doubles yearly. The stack uses Erlang/OTP with Riak and RabbitMQ and side‑effect functions for incremental identification and real‑time credit scoring. Half the new platform is live, shipped during peak season, and demonstrates practical gains from eventual consistency.
AI-written summary · about
chapters (21)
- 0:00Intro
- 1:23The Vision
- 2:47Payment Options
- 3:18Availability
- 4:09User Experience
- 4:53Migration Study
- 5:22Lean No Moisture
- 5:41The Challenge
- 6:21Side Effects
- 9:17Payment Gateway
- 11:14Synchronization
- 17:39Visualization
- 20:36Buffers
- 21:45Technical vs Business
- 23:45Risk Decisions
- 25:29Down Times
- 27:14Sticky Sessions
- 28:25Backend Communication
- 29:08PostgreSQL
- 29:46Open Source
- 30:22GUI
related talks
25mScaling AI for high-stakes, real-time payments
21mTransforming Legacy Banking Systems through Machine Learning
23mHow we unified feature engineering across data and backend
25mScaling Innovation: Rakuten’s Cloud-Native & AI Journey with
31mKlaviyo's Journey to Robust Model Serving with Ray Serve
54mKlaviyo Customer Agent: Building Bi-Directional Conversational Commerce at Scale
31mPursuing innovation with disruptive technologies: A fireside chat
22mData Ecosystem to power AI