Enhance Search Performance At Scale with
summary
Blinkit runs OpenSearch for 1M+ daily customers with ~750k queries/min peak and ~250k writes/day on a 30-data-node cluster holding a ~50GB product index across 2 primary shards and many replicas. Schema changes cut index size 48GB→32GB and reduced indexing latency ~6ms→1ms. Introducing batch writes slashes write disk IOPS and cuts write latency 15–20ms→2–3ms. Scaling nodes in multiples of three evens shard traffic and removes client timeouts.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:04Blinkit overview and scale
- 3:08OpenSearch at Blinkit
- 5:41High latency and update challenges
- 6:45Schema optimizations
- 10:00Architecture optimizations
- 15:18Batching real-time events
- 17:47Key learnings and conclusion
related talks
41mScaling Search and Powering Vector-Based Retrieval With OpenSea...
31mMaximize Resource Efficiency with Separated Index and Search Workloads
32mUnleash Your Cluster's Potential with OpenSearch Benchmark
38mScaling an online search engine to thousands of physical stores
36mScaling Log Management: PB-Scale Migration From Elasticsearch To...
33mLessons From Migrating To OpenSearch: Shard Design, Log Ingestion, and UI Decisions
38mMigrating from Self-Managed ElasticSearch to OpenSearch
36mSearch on the fly: how to lighten your Big Data