Prometheus

About Prometheus

Prometheus is an open-source systems monitoring and alerting toolkit, originally built at SoundCloud in 2012 by Matt T. Proud and Julius Volz, and donated to the CNCF in 2016 as its second graduated project. Prometheus's data model stores all metrics as time series identified by a metric name and key-value label pairs — enabling multi-dimensional data collection and powerful label-based filtering and aggregation. Prometheus's pull-based architecture is its defining design choice: Prometheus actively scrapes metrics from HTTP endpoints (/metrics) exposed by instrumented services and exporters, instead of services pushing to a central collector. This makes Prometheus configuration-centric rather than agent-centric, and enables health checking (a scrape failure means the target is down). PromQL (Prometheus Query Language) is a powerful functional query language for slicing and aggregating time series: rate(), histogram_quantile(), topk(), and label_replace() enable sophisticated metric analysis. Prometheus's Alertmanager handles alert deduplication, grouping, silencing, and routing to Slack, PagerDuty, OpsGenie, email, and webhooks. The Prometheus ecosystem includes 800+ exporters for databases, message queues, hardware, cloud APIs, and third-party services that expose metrics in Prometheus format. Prometheus Operator simplifies Kubernetes deployment via CRDs (ServiceMonitor, PodMonitor). The kube-prometheus-stack Helm chart bundles Prometheus, Alertmanager, and Grafana for a complete Kubernetes monitoring stack. Prometheus's limitations include no long-term storage (data is local, typically 15-day retention) — teams add Thanos, Cortex, or Grafana Mimir for multi-cluster, long-retention scalability.