Building a Time-Series Database for Point-in-Time Lookups
summary
G-Research builds a time‑series store on Aerospike using Aerospike CDTs and chunked monthly keys to hold compact, versioned chunks; temporal versioning answers “what was visible at time T” by storing a current map entry plus older chunk entries. Writes perform an insert-and-update within one generation using an atomic operate call so live systems get the latest value with low latency. Migration is incremental, multi-site is active‑active, upgrades show zero downtime, and the deployment doubles the variety of ingested series.
AI-written summary · about
chapters (21)
- 0:00Introduction
- 1:10Welcome
- 1:37Data is Everywhere
- 2:26Research
- 3:48Aims
- 4:49Chunks
- 5:35PointinTime
- 7:10Single Timestamp
- 8:54Get by Key
- 9:47Insert and Update
- 10:50Technology Selection
- 12:04Team Experience
- 13:46Business Benefits
- 15:59After a Year
- 19:13TimeSeries Database
- 20:46Compression
- 21:59Relational vs KeyValue
- 24:34TimeSeries APIs
- 26:29Expanding Data
- 28:03Increasing Sampling Rate
- 28:46New Feature
related talks
48mThe real-time database to build your AI future
39mTimescaleDB: Re-engineering PostgreSQL as a time-series database
48mRebuilding a RT Datastore: the design, deployment & performance story
30mCatching Waves with Time-Series Data
51mPostgreSQL, The Time-Series Database You Want
32mBuilding a Real-time Ingestion Platform using Kafka and Aerospike
42mAccessible Time Series data
08mBuilding a Real-time Data Future