Evolving Search Architecture for a Demanding Tesco Online Business
summary
Search architecture migrates to Kafka on Kubernetes with a Strimzi-like operator (called streamy in the slides), spinning a three-broker cluster in about three minutes and co‑housing namespaces for low latency. Elastic is upgraded from 5.x→8.x, cutting storage by ~30% and improving query latency; licensing cost per memory is reduced by running shards in pods. Data freshness targets ~5 minutes via feeder→enricher→Kafka→indexer pipelines. Vector Search A/B tests show more relevant results for queries like “kids snacks” and “cans of pop,” enabling a Search-as-a-Service platform others use.
AI-written summary · about
chapters (15)
- 0:00Introduction to Tesco search
- 1:27Search as art and science
- 2:09Business drivers for change
- 3:34Core architectural principles
- 4:59Elastic upgrade and Kubernetes shift
- 6:25Vector search architecture
- 9:58Vector search demo results
- 11:02Data liveliness pipeline
- 13:09Monitoring and observability
- 14:15Kafka platform for others
- 16:05Kubernetes operator benefits
- 17:09Search as a service
- 18:34Key learnings and risks
- 19:35Cost and simplicity focus
- 21:02Q&A
related talks
49mReplatforming Online Search for a Major High Street Retailer
21mHow Apache Kafka® Generates Value In Retail With Confluent
32mUnleashing the Power of Unstructured Data
28mNavigating the Transition: SRE Challenges and Highlights
36mModernizing Nubank's Kafka Platform for the Next 10 Years Starting at...
27mOptimize Your API with Event-Driven Microservice Architecture.
51mStreaming Platforms in Event-Driven Systems: Patterns, Pitfalls, and Performance
49mFrom research to production at ebay