VictoriaMetrics

VictoriaMetrics stores and queries time series data, most often as long-term storage for Prometheus or as a drop-in replacement for Prometheus and Graphite behind Grafana. It is optimized for the awkward case where old series are constantly retired and replaced by new ones at a high rate.

Two deployment shapes are open source under Apache 2.0: a single node and a cluster version. The single node has no dependencies, ships as one small binary, is configured through command-line flags, and the project argues it can stand in for medium-sized clusters built on Thanos, M3DB, Cortex, InfluxDB or TimescaleDB.

Ingestion and querying accept a wide spread of formats.

• Query languages: PromQL, and MetricsQL, the project's own faster dialect of it.
• Ingestion protocols: Prometheus remote write and exposition format, InfluxDB line protocol, Graphite, OpenTSDB, OpenTelemetry, DataDog, NewRelic, JSON, CSV and a native binary format.
• Stream aggregation: aggregate as data arrives, which lets it take the place of StatsD.
• Global query view: many Prometheus instances write into one store and are read with a single query.
• Operations: instant snapshots for backup and restore, metrics relabeling, a cardinality limiter and NFS-based storage such as Amazon EFS.

Published benchmarks claim up to seven times less RAM than Prometheus, Thanos or Cortex, and heavier compression than TimescaleDB, with Grammarly, Roblox, Wix and Spotify among the case studies. An enterprise edition adds anomaly detection, downsampling, multiple retentions and backup automation.