Scaling LinkedIn’s search infrastructure
summary
LinkedIn replaces legacy S E + Galene verticals with a managed hosted search platform that centralizes index builds, live updates and serving, cuts per-team deployables into lightweight artifact MPs, and loads them at runtime via the layer cake machinery with private classloaders. The system handles tenant index endpoints, blue‑green deploys, and an Espresso global secondary index flow, supports 170+ onboarded use cases, and operates at ~3.2M peak QPS, reducing operational cost and improving dev velocity.
AI-written summary · about
chapters (11)
- 0:00LinkedIn search scaling intro
- 1:34Core search infrastructure
- 4:04Legacy search challenges
- 7:56Tightly coupled search problems
- 9:25Hosted search overview
- 13:17Tenant index separation
- 15:26Runtime dependency management
- 17:53Global secondary index in Espresso
- 20:22Blue-green deployments
- 22:49Hosted search benefits
- 23:32Key learnings
related talks
37mScaling a Distributed Stateful System: A LinkedIn Case Study
49mReplatforming Online Search for a Major High Street Retailer
42mBuilding a real-time news search engine #bbuzz
24mReimagining Search : Scaling OpenSearch Across an Enterprise - Platform C...
44mSearching through large graphs using Elasticsearch
52mWhat’s New at SingleStore: The Latest Innovations
54mAdvanced (elastic)search for your legacy application
37mBuilding a Search Engine to Improve User Engagement