Prometheus: the monitoring tool that comes to you

Most old-school monitoring worked by push: each server sent its stats to a central collector. Prometheus, an open-source monitoring toolkit built at SoundCloud in 2012 and now a Cloud Native Computing Foundation project, flipped this. It uses a pull model — Prometheus periodically scrapes an HTTP endpoint (typically /metrics) that each app exposes, reading a simple text list of current numbers. Because the server initiates every request, it always knows exactly what it's trying to reach, so a target that goes silent is itself a signal that something is down.

Everything Prometheus stores is a time series: a metric name plus a timestamp, plus optional key-value labels like method="GET" or status="500". Labels are the secret sauce — instead of inventing a new metric for every combination, you attach dimensions and slice them later. A single http_requests_total series can be broken down by route, method, or status code on demand.

You query it with PromQL, a language built for math over time. You can ask for the per-second request rate over the last five minutes (rate(http_requests_total[5m])), aggregate across all instances, and feed the result straight into alerting rules or a Grafana dashboard. This label-plus-PromQL combo is a big reason Prometheus became the de facto standard for cloud-native metrics.