Netflix Boosts Druid Cache Hit Rate to 84%

Interval Caching: A Druid Game-Changer

Netflix's implementation of interval-aware caching in Apache Druid represents a sophisticated solution to a pervasive problem in real-time analytics: the inefficiency of traditional caching for rolling window queries. The core innovation lies in decomposing query results into time-aligned segments, allowing for the reuse of historical data. This approach directly tackles the redundant computation that plagues dashboards with frequently updating, overlapping time ranges. The reported 84% cache hit rate and 33% reduction in query load are substantial indicators of its effectiveness. The separation of query structure from time intervals and the use of a distributed key-value store for cached segments are sound architectural choices, enabling independent expiration and efficient retrieval. This strategy is particularly valuable for organizations operating at Netflix's scale, where trillions of rows and continuous monitoring demand extreme optimization.

However, the current deployment as an 'experimental layer' suggests that further refinement is ongoing. The reliance on an external proxy layer, while functional, adds architectural complexity. The future goal of integrating this caching mechanism directly into Apache Druid is a logical next step to streamline the architecture and potentially further optimize query planning. Extending support to templated SQL queries is also crucial for broader adoption within typical dashboarding tools, which often abstract away native Druid query expressions. While the article highlights significant performance gains, a deeper dive into the specific exponential TTL policies and the trade-offs involved in balancing data accuracy with freshness for different use cases would have provided even greater technical depth. The impact on P90 query times (a 66% improvement) is a powerful metric, but understanding the implications for other latency percentiles would offer a more complete picture of the performance benefits.

Key Points

• Netflix has implemented an interval-aware caching strategy for Apache Druid.
• This strategy significantly improves query efficiency for rolling window dashboards by reusing cached historical data.
• The system achieves an 84% cache hit rate and reduces query load on Druid by approximately 33%.
• Key innovation involves decomposing query results into time-aligned segments for reuse across overlapping queries.
• Cached segments are stored in a distributed key-value system with exponential TTL policies.
• The current implementation uses an external proxy layer, with future plans for direct integration into Druid.
• The solution offers substantial performance benefits, including reduced result bytes and segment scans, and improved P90 query times.

---

📖 Source: Netflix Serves 84% of Query Results from Cache with Interval-Aware Caching in Apache Druid