A single Prometheus instance has real limits — local disk retention, single-cluster scope, and memory pressure from high-cardinality metrics. As a kube-prometheus-stack deployment grows past one cluster or past a few weeks of retention, those limits start to matter.
This category covers the scaling path: integrating Thanos or Cortex for unlimited, object-storage-backed retention, running Prometheus federation or remote-write across multiple clusters, and the cardinality-control techniques that keep a growing deployment from running out of memory.
Configure remote_write in kube-prometheus-stack to ship metrics to Grafana Mimir, VictoriaMetrics, or another long-term storage backend — and when remote_write beats Thanos.
Read ArticleIntegrate Thanos with kube-prometheus-stack for unlimited metric retention and multi-cluster Kubernetes monitoring. Complete setup guide with S3, GCS, and production architecture.
Read ArticleUsually when local disk retention needs to exceed 30 days, when you need to query across more than one cluster, or when cardinality growth starts causing memory pressure — any of these is a signal to look at Thanos or remote_write.
No — remote_write to a managed backend like Grafana Mimir is a simpler alternative when a centralized backend already exists; Thanos is better when you want to keep everything within your existing Prometheus/Grafana setup.
