Prometheus Fundamentals

Prometheus answers a practical operations question: what is your system doing now, and how has that behavior changed?

It collects numeric measurements called metrics. Each measurement becomes a sample with a value and timestamp. Prometheus stores related samples as a time series, then lets you select and combine those series with PromQL.

This model works well for request counts, error rates, latency, queue depth, memory use, and other numeric signals. It does not preserve the complete detail of every event. Use logs or traces when you need individual event context.

The core mental model

Think in one repeating loop:

1. A service or exporter exposes metrics over HTTP.
2. Prometheus discovers a target and scrapes its metrics endpoint.
3. Prometheus stores each sample in its local time series database.
4. You use PromQL to select, filter, aggregate, and compare time series.
5. Recording rules save useful query results as new time series.
6. Alerting rules identify conditions that need attention.
7. Alertmanager groups, routes, silences, and inhibits notifications.

Prometheus usually pulls metrics from targets. This makes the server responsible for scrape timing and target health. The Pushgateway covers a narrower case: short-lived jobs that may finish before Prometheus can scrape them.