What is a time-series database (TSDB)

A time-series database (TSDB) is a database optimized for storing and querying time series: sequences of values with timestamps. Monitoring is built on them — every metric is a series of "time → value" points.

How a TSDB differs from a regular database

• Data is almost always appended, not updated — which allows aggressive compression.
• Queries are almost always over a time range with aggregation: "average CPU across instances over the last hour".
• Old data gets cheaper: it is downsampled or dropped by retention.

Key concepts

• Series — a unique combination of a metric name and a label set, e.g. http_requests_total{job="api", code="500"}.
• Cardinality — the number of unique series. High cardinality (e.g. a user_id label) blows up memory and cost.
• Scrape / ingest — the interval at which new points arrive.
• Retention — how long raw points are kept.

Examples of TSDBs

Prometheus, VictoriaMetrics, Thanos, Mimir, InfluxDB, TimescaleDB. Unimoni uses a Prometheus-compatible store, so queries are written in PromQL.

sum(rate(http_requests_total[5m])) by (code)

What matters in practice

Watch cardinality: do not put high-variance values (user_id, request_id, full URL) into labels. That is the main reason monitoring becomes "expensive" and slow.