The de facto standard for cloud-native metrics — a pull-based time-series database paired with PromQL and an alerting rule engine. The data model (labeled metrics with arbitrary key-value dimensions) and the operator pattern on Kubernetes have made it the metrics layer that almost every other tool integrates with, including Datadog, New Relic, and Grafana. Remote-write support means you can keep the scraping topology and ship metrics to a long-term backend like Mimir, Thanos, or Cortex without rewriting your scrape config.
High fit for any team running Kubernetes — the kube-prometheus-stack Helm chart gives you cluster, node, and workload metrics within an hour, and PromQL is the lingua franca of every SRE interview. It is also the right choice when regulatory or contractual rules forbid sending metrics to a SaaS vendor.
Watch out for: Prometheus on its own is a single-node store with retention measured in weeks, not years. Production deployments at scale require a distributed backend (Mimir, Thanos) and the operational know-how to run it. High-cardinality labels — user IDs, request IDs, anything unbounded — will OOM your Prometheus server in minutes if you are not careful.
