PCA: The Prometheus Certified Associate Exam, Domain by Domain

The Prometheus Certified Associate is the CNCF and Linux Foundation credential for engineers who monitor systems with Prometheus. Unlike the CKA or CKAD it is not performance-based: it is an online, proctored, multiple-choice exam lasting 90 minutes, priced at $250 for the exam alone, and the certification is valid for two years. The Linux Foundation does not publish a question count or a pass mark, so treat any number you see for those as hearsay.

What it does publish is the blueprint, and CNCF open-sources the curriculum itself, so the domain list is not something a prep site has to guess at. Five domains, and the weights are the whole study plan: PromQL 28 %, Prometheus Fundamentals 20 %, Observability Concepts 18 %, Alerting & Dashboarding 18 %, Instrumentation and Exporters 16 %.

PCA is one of eight CNCF associate exams, all priced and shaped alike — if you are still deciding between them, which cloud native certification should you take? compares all eight side by side. Gnoseed does not sell, proctor or accredit this exam and has no affiliation with CNCF or the Linux Foundation. What it offers is free spaced-repetition practice on the material the blueprint names — so verify every fact above against the official exam page before you book, because prices and formats move.

If you rank the domains by weight, PromQL is the largest by eight points, and it is also the part that punishes recognition-level knowledge hardest. Knowing that rate() exists is not the same as knowing what it returns when a counter goes from 100 to 400 across a five-minute window, why rate(sum(x)) is not sum(rate(x)), or why a topk(3, …) graph can draw more than three lines.

The curriculum names seven PromQL competencies: selecting data, rates and derivatives, aggregating over time, aggregating over dimensions, binary operators, histograms, and timestamp metrics. Two of those trip people who use PromQL daily. Aggregating over time (avg_over_time and its family) is a different operation from aggregating over dimensions (avg, sum by) — one collapses a range vector per series, the other collapses across series at one instant. And timestamp metrics — a metric whose value is a Unix time — are read with time() - metric, which is obvious only once you have seen it.

A practical rule: if your revision time is limited, spend it in blueprint order. Perfecting Alertmanager routing trees buys 18 % of the paper; being fluent in PromQL buys 28 % and makes the Alerting domain easier as a side effect, because alerting rules are PromQL expressions.

The fifth domain is Observability Concepts at 18 %, and its published competencies are: metrics, understanding logs and events, tracing and spans, push versus pull, service discovery, and the basics of SLOs, SLAs and SLIs. Read that list again — tracing, logs and service-level objectives are not Prometheus features. Prometheus stores numeric time series and nothing else; traces live in Tempo or Jaeger, logs in Loki or Elasticsearch, and SLOs are an SRE practice that Prometheus merely happens to be good at measuring.

That is why a candidate who has only ever operated Prometheus can lose most of a domain worth roughly a fifth of the exam. You are expected to know what a span is and how it relates to a trace, why logs answer a question metrics cannot, and how an SLI differs from an SLO and from the contractual SLA sitting behind it — without any of it being a question about a scrape config.

It also explains the study plan on this page. The Prometheus deck covers the four Prometheus-shaped domains and the SLI side of the fifth; the Observability deck covers signals, traces, spans and the OpenTelemetry vocabulary; and the SRE deck covers SLOs, SLAs, error budgets and the when-what-why of alerting. Three decks, because the blueprint genuinely spans three subjects.