TLDR Tech / DevOps
Stop Answering the Same Question Twice: Interval-Aware Caching for Druid at Netflix Scale
Netflix built an experimental caching layer for its Apache Druid database that now handles over 10 trillion rows and ingests up to 15 million events per second, reducing query load by 33% and improving response times by 66%. The system splits time-series results into independently cached buckets with age-based TTLs (5 seconds for fresh data, up to 1 hour for older data), allowing dashboards to serve 84% of result data from cache while only querying Druid for the newest data—a critical improvement during high-profile live events when dozens of engineers monitor the same dashboards simultaneously.
Read the full story