Observability Wiki
Open source

Grafana Mimir

Grafana Labs' horizontally scalable, long-term storage backend for Prometheus metrics, built as the successor to the Cortex project.

Last updated

What it is

Grafana Mimir is an open-source, horizontally scalable time series database purpose-built as a long-term remote-write backend for Prometheus-compatible metrics. Grafana Labs released it in 2022 by forking Cortex — a CNCF project Grafana Labs had been a lead contributor to for years — under the AGPLv3 license, combining lessons learned running Cortex at scale for Grafana Cloud customers with a redesigned, more operationally simple architecture. Since the fork, Cortex development has slowed considerably and most of its former maintainers now work on Mimir, effectively making Mimir the active successor for teams that want Cortex-style scalable Prometheus storage going forward. Mimir supports multi-tenancy, horizontal scaling of ingestion and querying independently, and (as of its 3.0 release) an optional architecture that uses Apache Kafka as a buffer between ingestion and querying to improve durability and decouple write and read paths.

Mimir is the metrics component of Grafana Labs’ broader open-source observability stack, alongside Grafana Loki for logs, Grafana Tempo for traces, and Grafana Pyroscope for profiling — all queried through Grafana itself.

Why teams choose it

  • Purpose-built for scale that vanilla Prometheus can’t reach alone — single-node Prometheus doesn’t scale horizontally or retain data indefinitely, and Mimir addresses both by design, at the cost of a materially more complex system to operate (many microservice components vs. one binary).
  • Multi-tenancy is a first-class feature, making it a natural fit for platform teams offering metrics-as-a-service to multiple internal teams, which simpler alternatives like VictoriaMetrics support less natively.
  • It’s also the engine behind Grafana Cloud’s metrics, so choosing self-hosted Mimir gets you the same technology Grafana Labs runs in production at very large scale — though running it well yourself requires real operational investment (object storage, careful capacity planning across components).
  • Migrating from Cortex is a supported, documented path, but teams still running Cortex should treat it as an eventual requirement rather than an option, given the maintenance-mode trajectory of the original project.

Pricing model

Mimir itself is free, open-source software under the AGPLv3 license, with no separate “enterprise” fork of the core engine. Grafana Labs’ commercial angle is Grafana Enterprise Metrics (added enterprise support, security features, and integrations on top of self-hosted Mimir) and, more commonly, Grafana Cloud, where Mimir runs as a fully managed service billed on usage (active series and ingested samples being the typical basis).

Alternatives

The most direct alternatives are other Prometheus remote-write compatible long-term storage systems: VictoriaMetrics (generally simpler to operate, lower resource footprint) and Thanos (an older, still-active project with a different architecture built around object storage and a sidecar model). Cortex itself remains an alternative in name only, given its diminished development activity. Outside the Prometheus ecosystem, commercial platforms like Datadog and Amazon CloudWatch offer managed metrics storage without any of this operational overhead.

Visit official site →