OpenSearch as the Unified Backend: From Log Storage to Agentic Search Platform at LINE
summary
LINE runs about 1,280 clusters across 10,000+ nodes and grows from log storage to a high‑performance search and vector backend powering core services. Engineers implement automatic multi‑AZ node distribution and cross‑region snapshots, achieve zero data loss, use MVM over Fabric for hot nodes and a Cold Tier plus remote snapshots for year‑plus retention, and tune hardware generations for better throughput. Roadmap adds hybrid search and vector search with GPU automation, plus cloud‑level ingest pipelines and automated model deployment and access control.
AI-written summary · about
chapters (7)
- 0:00LINE’s OpenSearch overview
- 2:27Why switch from Elasticsearch
- 4:15OpenSearch architecture at LINE
- 7:02Data tiering and scaling
- 8:28Performance tuning and hardware
- 9:31OpenSearch adoption in LINE services
- 10:37AI agent support and roadmap
related talks
25mKeynote: OpenSearch in the Agentic Era
26mOpenSearch: Conversations with Context
06mSponsored Keynote: OpenSearch - See Everything: Open Observability for Agentic AI
41mKeynote: From Keyword Search to Vector Powered Insights: Th...
47mBoosting AI With OpenSearch: Vector Stores, Similarity Search, and Retrieval-Augme...
40mBuilding Intelligent Search Systems With Agentic RAG and OpenSearch
42mObserve Smarter, Not Harder: Scaling AI-Powered Observability With O...
42mOpenSearch With Gaia: Building Next-Gen AI-Driven Search and Analytics Workflows