Grafana Mimir

Grafana Mimir is an open-source, horizontally scalable long-term storage system for Prometheus metrics, developed by Grafana Labs and announced in 2022. It addresses the fundamental limitation that Prometheus stores metrics only on local disk, with retention limited by single-node storage capacity. Mimir provides a distributed backend that enables Prometheus to scale from days of retention to years, and from millions of active time series to billions.

Mimir's architecture separates ingestion, query, and storage into independently scalable components. Ingesters accept metrics from Prometheus remote-write endpoints, compact them into blocks, and store them in object storage such as Amazon S3 or Google Cloud Storage. Queriers execute PromQL queries across distributed blocks, using a store-gateway to locate relevant data. The result is a Prometheus-compatible query interface with the durability and scale of cloud-native object storage.

The project emerged from Grafana Labs' earlier Cortex project, which it forked and simplified to reduce operational complexity. Mimir represents a specific theory about observability infrastructure: that the metrics store should be a commodity service, as generic as a filesystem, and that the value lies in the query and visualization layers above it. This theory is not uncontroversial. It implies that metrics data is interchangeable, that the query language is the interface that matters, and that organizations should not build custom metrics storage. For organizations whose metrics encode proprietary business logic, this commodification assumption may be wrong.