Time series database

A time series database (TSDB) is a database optimized for storing and querying data indexed by time — sequences of timestamped measurements, events, or states. Unlike general-purpose relational databases, which treat time as an ordinary attribute, a TSDB builds its storage engine, indexing strategy, and query optimizer around the assumption that temporal ordering is the primary access pattern. This specialization enables ingestion rates and compression ratios that relational databases cannot match, at the cost of abandoning the relational model's flexibility: joins, foreign keys, and complex transactional semantics are typically absent or severely constrained.

The time series database landscape is fragmented and philosophically diverse. Prometheus optimizes for operational metrics in containerized environments, using a pull model and dimensional labeling. InfluxDB optimizes for high-throughput event streams and IoT data, using a push model and tag-based metadata. TimescaleDB attempts a hybrid approach by extending PostgreSQL with time-series optimizations, preserving SQL compatibility at the cost of some performance. Apache Druid and ClickHouse target analytical workloads on massive historical datasets. Each of these systems encodes a different theory of what time-series data is, how it arrives, and what questions users will ask of it.

The unsolved problem at the center of the TSDB field is the tension between real-time and historical analysis. A system optimized for millisecond-latency ingestion is not the same system optimized for decade-spanning analytical queries. The architectures that solve one problem well — in-memory buffering, LSM trees, columnar Parquet files — are poorly suited to the other. The time series database is therefore not a single category but a spectrum, and the choice of database is a choice about what kind of time matters most.