Prometheus

In 2012, SoundCloud had a growing microservices architecture and no good way to monitor it. Existing tools were designed for monolithic applications on physical servers. They needed something that could handle dynamic service discovery, high-dimensional metrics, and the ephemerality of containers.

Two ex-Google engineers. Matt Proud and Julius Volz, built Prometheus, inspired by Google internal monitoring system Borgmon. It was open-sourced in 2012, joined the CNCF in 2016 as the second hosted project (after Kubernetes), and is now the de facto standard for metrics monitoring in cloud-native environments.

Prometheus is not just popular, it defined how the industry thinks about metrics. The metric naming conventions, the label-based data model, the pull-based scraping architecture, and the PromQL query language have influenced every monitoring tool built since.

Prometheus uses a pull model: it periodically scrapes HTTP endpoints exposed by your services to collect metrics.

Core architecture:

• Targets. Your services expose a /metrics endpoint with current metric values in Prometheus text format
• Scraping. Prometheus pulls from each target every 15-60 seconds (configurable)
• TSDB. A local time-series database stores the scraped data, optimized for append-heavy workloads with efficient compression
• PromQL. A purpose-built query language for slicing, aggregating, and transforming time-series data
• Alertmanager. A companion service that evaluates alert rules, deduplicates, groups, and routes notifications

Service discovery: Prometheus integrates with Kubernetes, Consul, AWS EC2, DNS, and file-based targets. In Kubernetes, it automatically discovers pods with specific annotations, no manual target configuration needed.