Time Series Databases

databasecompanysqlrelease

Stores optimised for timestamped, append-mostly data: metrics, events, sensor readings. The distinguishing features are cheap ingestion of ordered writes, downsampling or retention policies, and query languages built around time windows rather than joins.

General purpose

  • InfluxDB - the widest-known of the dedicated engines.
  • TimescaleDB - a PostgreSQL extension rather than a separate server, so timestamped data stays in the database you already run. See PostgreSQL. Split licensing: LICENSE says the source is “variously licensed under the Apache License Version 2.0, an Apache compatible license, or the Timescale License”, so the community features are source-available rather than open source. The company renamed itself TigerData on 17 June 2025, which is why search results disagree about what it is called.
  • QuestDB - a “column-oriented, time-partitioned storage engine with memory-mapped files and vectorized (SIMD) execution”, with SQL extended by SAMPLE BY, LATEST ON and ASOF JOIN.
  • VictoriaMetrics - describes itself as “long-term storage for Prometheus, or as a drop-in replacement for Prometheus and Graphite in Grafana”. Apache 2.0, with separate enterprise releases.
  • Apache Druid - real-time analytics rather than pure metrics.
  • KairosDB - “written on top of Cassandra”.
trial

TimescaleDB, for staying inside Postgres.

trial

InfluxDB, when a dedicated server is genuinely warranted.

Monitoring stacks

  • Prometheus - a monitoring system that happens to contain a time series database; the pull model and alerting are the point. See Prometheus, Alertmanager and Grafana.
  • Graphite - the older graphing stack, split across Whisper (the file format) and Carbon (the ingestion daemon).
  • RRDtool - the original round-robin database. Fixed-size files, automatic downsampling, still behind a lot of network monitoring.

Slowing down

  • OpenTSDB - on HBase. Last release activity is well behind the others here.

Related