Time-series database
A database built to store and query timestamped measurements, which is what keeps monitoring charts fast over long time ranges.
See it
What it is
A time-series database, or TSDB, stores repeated measurements indexed by time: CPU usage every 15 seconds, requests per minute, temperature each hour. It groups points into named series, usually one series for every unique set of labels, then compresses adjacent timestamps and values so range scans and time-window aggregations stay fast. Prometheus includes a TSDB; InfluxDB and TimescaleDB are other familiar examples.
Reach for one when the main questions are 'what changed over this window?', 'what is the rate?', and 'when did this cross a threshold?' Retention rules and downsampling let you keep recent points at full resolution while preserving older history as coarser summaries.
Gotcha: the expensive unit is often the series, not the individual point. Adding an unbounded label such as user_id or a raw URL creates a new series for every distinct label combination, consuming index memory even when each series contains few samples. A TSDB is excellent at dense numeric history and a poor substitute for searchable request logs.
Ask AI for it
Set up Prometheus as the time-series database for this service. Scrape its /metrics endpoint every 15 seconds, attach only service, environment, route template, and status class labels, and retain raw samples for 30 days. Add recording rules for five-minute request rate, error rate, and p95 latency so dashboards do not repeat expensive PromQL. Configure remote_write to Grafana Mimir for long-term storage, and create an alert on active series count so a label-cardinality regression is caught before it exhausts memory.