TLDR Tech / Data
Stop Answering the Same Question Twice: Interval-Aware Caching for Druid at Netflix Scale
Netflix built a caching layer in front of Apache Druid to stop answering the same time-series queries by intercepting queries at the Druid Router, parsing the query structure, and storing results in fine-grained time buckets using a Cassandra-backed cache. For overlapping windows, it serves cached data for settled intervals and only fetches the missing recent tail from Druid. It uses exponential TTLs and gap-aware merging to balance freshness with cache hit rates.
Read the full story