Scaling Time-Series Data to Infinity: A Kubernetes-Powered Solution with Envoy
summary
Built an object-storage backed time‑series store on Kubernetes using S3-compatible buckets, Envoy for consistent‑hashed L7 routing, and a two-layer design with LevelDB aggregators and Bolt/LDB caches. System handles 1 billion metrics and ~1 PB of samples with 2.7 TB/day ingestion; Cassandra at 420 nodes is replaced for cold data. 32 shard pods aggregate 450 GB every 4 hours in ~40 minutes using merge files plus an index-of-index to cut CPU and network, and read p99 latency becomes comparable to Cassandra.
AI-written summary · about
chapters (10)
- 0:00Scaling time-series data
- 1:07Petabyte-scale metrics challenges
- 5:51Cassandra bottleneck
- 9:38Object storage for time-series
- 12:08Data structure design
- 15:51Kubernetes for scalability
- 20:44Reading from object storage
- 25:07Query performance optimization
- 27:23Bring your own bucket
- 28:30Conclusion
related talks
07mLightning Talk: Scaling Time Series Databases on Kubernetes
14mBuilding the future of time series data
34mHow Netflix Delivers Key-Value and Time-Series Storage at Any Scale
30mCatching Waves with Time-Series Data
39mA New Data Model for Time Series Data Processing
49mThinking in time series, the new solution to your pro...
32mDiscover the Modern Way to Build Real-Time Analytics Solutions at Any Scale
29mCatching Waves with Time-Series Data