Open heart surgery - Replacing our search index on the fly
summary
Replaces a Solr 5.3 index serving ~2 million queries/day and 1.5M weekly users with a new Solr 9.x deployment on Azure, moving ~10 on‑prem Red Hat servers while keeping read availability. Key moves are copy index files from stage to prod after making stage mirror production, using Zookeeper, refeeding largest collections slowly, and doing a rolling replace of URLs for swapover. Migration takes 2.5 years end‑to‑end, yields zero read downtime and only a few hours of planned write maintenance.
AI-written summary · about
chapters (9)
- 0:00Introduction and background
- 3:36The Solr setup and challenges
- 6:27Project escalation and complexity
- 10:22Technical debt and refeeding dilemma
- 13:55Data quality and feeding strategies
- 16:48Testing and staging environment
- 19:12Going live with zero downtime
- 21:42Lessons learned and advice
- 23:06Closing and inspiration
related talks
30mUsing Solr unconventionally to serve 26bn+ documents
34mArchitecting Solr indexing pipelines in Google Cloud Platform
34mEvolution of Slack Search
46mSearch in the Cloud: separation of compute and storage
19mYou Know, for Search. Querying 24 Billion Records in 900ms
53mA cloud migration story - running at 100% all while migrating and shipping new features
38mScaling an online search engine to thousands of physical stores
36mWhat’s New in the OpenSearch Project and Ecosystem