Prometheus

Prometheus is an “open-source systems monitoring and alerting toolkit” (SoundCloud origin; the second CNCF project after kubernetes, 2016). It’s the de-facto metrics backend of the CNCF stack this spoke keeps meeting — the very tool the kubernetes-integration-tax piece names (Cilium metrics invisible to it without a ServiceMonitor). Source: official Prometheus docs.

Model

• Multidimensional time series identified by “metric name and key/value pairs” (labels);
• pull-based “collection happens via a pull model over HTTP” (scrape), with a push gateway for ephemeral jobs;
• PromQL query language; Alertmanager for alerting; each server is standalone (no network storage dependency — reliability under failure).

Strengths vs. limits

Great for “purely numeric time series” in microservices; explicitly not for billing-grade data (“not be detailed and complete enough … for 100% accuracy”) — the same metrics-vs-exactness caveat worth recording for observability. Pairs with opentelemetry (which can produce/ship metrics into a Prometheus-compatible store).

Freshness

Re-read against the same overview page 2026-08-12: every claim above still holds word for word — the SoundCloud origin, “the second hosted project, after Kubernetes” in 2016, the dimensional model, PromQL, and the billing-accuracy exclusion. The page names no version number and no OTLP ingestion, so neither is recorded here; if OTLP support lands in this overview it is the change to catch, since it would blur the OTel-produces / Prometheus-stores split this page draws.

Related

observability · kubernetes · kubernetes-integration-tax · opentelemetry · platform-ops · site-reliability-engineering