Time-series data is data measured over time.

Servers emit CPU and memory metrics. Applications record request rates and latency. Sensors report temperature. Exchanges publish prices and trades. All of these records have one thing in common: the timestamp is central to how the data is stored, queried, and expired.

Time-series databases are optimized for this shape of workload. Writes are mostly append-only and queries usually include a time range.

Aggregations over windows are more common than single-record lookups, and recent data is usually queried more often than old data. Old data often expires or is downsampled to a lower resolution.

A regular relational database can handle modest time-series workloads with good indexes and partitioning. A dedicated time-series database becomes useful when ingest volume, retention, compression, rollups, or cardinality become central design concerns.