**Mimir** is Grafana Labs' open source, horizontally scalable, long-term metrics storage system that is fully compatible with [[Prometheus]]. It stores Prometheus metrics at scale and provides a multi-tenant, globally queryable endpoint — replacing [[Thanos]] for new deployments that prefer a more integrated, operationally simpler approach.
---
### First Principle: Long-term metrics storage should scale horizontally, not by buying larger disks.
[[Thanos]] achieves scale through composition of existing Prometheus instances. Mimir achieves it through a purpose-built, horizontally scalable microservice architecture — each component (ingester, querier, store-gateway, compactor) scales independently based on load.
---
### Key Considerations
- **Multi-Tenancy**: Mimir is designed for multi-tenant environments from the ground up — every metrics query and write is tenant-scoped.
- **Prometheus-Compatible**: Mimir accepts data via Prometheus remote_write and serves queries via the Prometheus HTTP API. Any tool that works with Prometheus works with Mimir without changes.
- **Microservice Architecture**: Decomposes into ingesters (buffer writes), distributors (route writes), queriers (execute queries), store-gateways (serve stored blocks), and compactors.
- **Object Storage Backend**: Mimir stores all metrics data in object storage ([[MinIO]], [[Ceph]], S3, GCS) — not on local disks. Storage cost is proportional to data volume.
- **vs [[Thanos]]**: Thanos is more modular (uses existing Prometheuses). Mimir is more integrated and operationally simpler for operators starting fresh.
---
### How It Fits
```
[[Prometheus]] instances (remote_write to Mimir)
→ Mimir (distributed ingest + storage on [[MinIO]])
→ Mimir Querier (PromQL at global scale)
→ [[Grafana]] (dashboards)
```
[[Prometheus]] | [[Thanos]] | [[Grafana]] | [[MinIO]] | [[Open Source Hyperscaler MoC]]