# Inject 5s latency into 50% of scrape requests for 2 minutes curl -X POST http://localhost:9091/inject/latency \ -d '"duration":"2m","percent":50,"delay":"5s"'

While it’s often considered the "ultimate" version for fans who wanted more answers, here is a guide to what makes it interesting: Key Additions & Changes The "Sacrificial" Elder Engineer:

is the practice of intentionally weaponizing Prometheus against itself. It involves injecting faults directly into the monitoring pipeline to verify that your observability survives the crash. If your metrics lie to you during a disaster, you are flying blind.

A successful test isn’t “nothing broke.” A successful test is: “We detected the anomaly, mitigated the blast radius, and fixed the root cause without user impact.”

: It provides a deeper context for the "black goo" and the Engineers' motivations, often drawing from supplemental materials not found in the standard Blu-ray. Why Fans Seek It

If you run Prometheus Operator, pair it with (CNCF project) and a NetworkChaos experiment:

apiVersion: chaos-mesh.org/v1alpha1 kind: NetworkChaos metadata: name: prometheus-slow-scrape spec: action: delay mode: all selector: pods: prometheus-ns: - prometheus-server-0 delay: latency: "3s" correlation: "100" jitter: "1s" duration: "5m"

Consider the "Silent Scrape Failure." Standard Prometheus metrics are pull-based. If your service is up but the /metrics endpoint returns a 500 error because the JSON marshaller is broken, Prometheus records a "down" sample. But is the service down? No. The monitoring is down.

Until then, Prometheus Chaos Edition remains a philosophy for the brave. It asks the hard questions: Is your monitoring reliable, or is it just lucky?

This creates fake CPU readings without touching the actual exporter.

Prometheus: Chaos Edition is a highly regarded fan edit by editor Studio Yutani that attempts to "fix" Ridley Scott's 2012 film by re-integrating nearly all deleted, extended, and promotional footage into a single, comprehensive narrative.

არქივი