Mimir

Why it's ranked here

It carries AGPL-3.0 licensing, a documented multi-maintainer governance process with formal voting, and a lineage traceable through code comments to Cortex and Thanos, the two projects it absorbed ideas and code from. That heritage shows in how much of the read path, like block verification and out-of-order chunk detection, was carried over nearly verbatim with attribution intact. A project with a formal governance charter and named release shepherds per version is built for institutional use, not a side project.

What's good

The storage tier accepts any S3-compatible backend, not just AWS, along with Google Cloud Storage and Azure Blob, so teams are not locked into one cloud vendor for the expensive part of the system. Multi-tenancy is built into the architecture rather than bolted on, with per-tenant limits enforced at the API layer. The block-repair tooling can detect out-of-order chunks, duplicate chunks, and unsupported index formats before they cause query errors, which is the kind of maintenance code most databases leave to operators to write themselves.

Tradeoffs

None of this comes free: running it as separate ingester, compactor, and store-gateway services means operating a distributed system with its own failure modes, not a single Prometheus binary. The license is AGPL-3.0-only, which some companies avoid for anything they might modify and run as a network service without contributing changes back. The command-line tool ships as a second binary with its own subcommands, so operators are learning two tools' worth of flags, not one.

How to use it well

It fits teams that have already outgrown a single Prometheus instance's retention or query load and need one queryable view across many source servers or tenants. Start in the single-binary mode to learn the configuration surface before splitting components out for independent scaling. Use the bundled CLI for config validation and historical backfills rather than hand-rolling those against the API. It is not a fit for a small setup that just wants Prometheus with longer retention; the multi-component design is overhead you pay for scale you may not need yet.

Observed

License

AGPL-3.0-only (full text present in LICENSE).

Language

Go, packaged as a single module.

Packaging

one binary that runs all components together in monolithic mode, or splits into independent services (ingester, compactor, store-gateway, query frontend, etc.).

Interfaces

a CLI server binary, a separate management CLI (mimirtool) with subcommands for access control, alerting, and config validation, and an HTTP API shaped after Prometheus's own query endpoints.

Structural

Ships a standalone comparison-proxy tool (query-tee) that mirrors read requests to two backends and diffs the responses.

Structural

Storage backend is pluggable object storage: AWS S3, Google Cloud Storage, Azure Blob Storage, OpenStack Swift, or any S3-compatible store.

Architecture

Multi-tenant by design, with tenant isolation and per-tenant limits built into the architecture.

Structural

The TSDB block index reader only recognizes index format v2.

Structural

Source files carry explicit provenance headers attributing shared components to the Cortex and Thanos projects.

Governance

Formal written governance (GOVERNANCE.md) defines team membership, maintainer roles, and a supermajority voting procedure.