Thanos

Thanos takes Prometheus and makes it work at scale. Long-term storage in object stores (S3, GCS), global query across multiple Prometheus instances, downsampling for historical data, and high availability. It's Prometheus without the "what happens when the disk fills up" problem.

If you're running Prometheus and hitting storage limits or need to query across clusters, Thanos is the standard upgrade path. Cortex solves the same problem with a different architecture (multi-tenant by design). Mimir from Grafana Labs is the newer option with better performance. Commercially, Grafana Cloud manages this stack for you.

The sidecar approach is elegant — add Thanos to existing Prometheus instances without changing your setup. Compaction and downsampling keep long-term storage costs manageable.

The catch: Thanos adds operational complexity on top of an already complex Prometheus stack. Object store configuration, compactor tuning, and query latency for historical data all need attention. If you're running a single Prometheus instance and it's working fine, you don't need Thanos yet. And Mimir is increasingly the recommended choice for new deployments in the Grafana ecosystem, making Thanos feel like the "previous generation" solution.