Here are 20 free PCA sample questions with answers, taken from the GoldenKube question bank and grouped by exam domain. Every question has the correct answer marked and a written explanation, so you can check what you know and see why the wrong options fail. Prometheus Certified Associate is a multiple-choice exam; this set is independent practice and not affiliated with the Linux Foundation.
How to use these PCA practice questions
Work through the 20 PCA sample questions in one sitting without looking anything up. For each question, decide on an answer first, then open the answer and read the explanation, including when you were right, because a correct guess and a correct line of reasoning look identical until the exam rewords the question. Mark every question as right, wrong or guessed, and count a guess as wrong when you score yourself. Give the set about 30 minutes, which is the average time per question on the real exam multiplied by twenty. The aim is to find out which exam domains are already solid and which ones need study before you buy the exam.
The official exam format is 60 questions in 90 minutes, with a passing score of 75%, according to the Linux Foundation documentation checked on 2026-10-23.
With JavaScript enabled, each answer starts collapsed behind a Show answer button so you can test yourself. Without JavaScript, and for any crawler or screen reader that reads the raw page, every answer and explanation is already visible below its question.
These questions were written for GoldenKube practice. They are not copied from the real exam, and nothing on this page tells you what the real exam will ask.
Read the question twice and underline the constraint words: "most", "least", "first", "only", "not".
Eliminate options you can prove wrong before you pick between the rest.
Where a question says to select all that apply, every correct option must be chosen and none of the wrong ones.
Read the explanation even for questions you got right; it often names the nearby concept the exam likes to swap in.
Observability Concepts (4 questions)
Question 1
What is an 'event' in observability terms?
A.A continuously increasing counter exposed by an exporter
B.A pre-aggregated rollup of metric data over a time window
C.A sampled gauge value representing current resource usage
D.A record indicating that something notable occurred at a specific point in time
Correct answer: D
An event captures the fact that something discrete happened at a moment in time (for example, a deployment or a login). Counters, gauges, and aggregated rollups are forms of metric data, not events.
Question 2
The three commonly cited 'pillars of observability' are:
A.Metrics, logs, and traces
B.Counters, gauges, and histograms
C.Alerts, dashboards, and reports
D.Push, pull, and scrape
Correct answer: A
Metrics, logs, and traces are the three classic telemetry signals used to understand a system's behavior. Counters and gauges are metric types, and the other options describe Prometheus features rather than observability pillars.
Question 3
What does a Service Level Indicator (SLI) represent?
A.A quantitative measure of some aspect of the level of service provided
B.A legal contract with financial penalties for downtime
C.The internal target a team commits to for reliability
D.A list of scrape configurations for a job
Correct answer: A
An SLI is the actual measurement, such as request latency or the ratio of successful requests. The target for that measurement is the SLO, and a contractual agreement is the SLA.
Question 4
What does a trace represent in distributed tracing?
A.The end-to-end path of a request as it propagates through multiple services
B.A single numeric sample taken at a regular interval
C.A configuration file describing scrape targets
D.An alert that fires when a threshold is crossed
Correct answer: A
A trace ties together the spans produced as a single request moves through a distributed system, giving an end-to-end view. Numeric samples are metrics, and scrape configuration and alerts are unrelated concepts.
Prometheus Fundamentals (4 questions)
Question 5
Which two labels does Prometheus attach automatically to scraped samples?
A.__name__ and le
B.job and instance
C.severity and team
D.quantile and bucket
Correct answer: B
When scraping, Prometheus automatically adds the job label (from the scrape job name) and the instance label (the target's host:port). The other labels are user-defined or specific to certain metric types.
Question 6
What does a # HELP line provide in the exposition format?
A.The current numeric value of the metric
B.A human-readable description of the metric
C.The scrape interval to use for the metric
D.A list of permitted label values
Correct answer: B
The # HELP line gives a short, human-readable description of what the metric measures. It is informational and does not affect collection or values.
Question 7
What is the primary job of the Prometheus server itself?
A.To render dashboards and host a Grafana instance
B.To scrape and store time series data and evaluate rules
C.To forward all metrics to a relational database
D.To push metrics to remote agents on a schedule
Correct answer: B
The Prometheus server scrapes targets, stores the samples in its local time series database, and evaluates recording and alerting rules. Dashboarding is typically delegated to Grafana.
Question 8
In the text exposition format, what does a # TYPE line do?
A.Declares the retention period for the metric
B.Sets a label value on the metric
C.Specifies the metric's type (counter, gauge, histogram, or summary)
D.Defines an alerting threshold for the metric
Correct answer: C
A # TYPE comment line tells Prometheus how to interpret the following metric (counter, gauge, histogram, or summary). It does not configure retention, labels, or alerts.
PromQL (6 questions)
Question 9
How does irate() differ from rate()?
A.irate() works only on gauges, not counters
B.irate() returns a range vector instead of an instant vector
C.irate() uses the last two data points for an instantaneous rate, while rate() averages over the whole range
D.irate() is identical to rate() but only valid for summaries
Correct answer: C
irate() computes an instantaneous rate from the two most recent samples in the range, making it responsive but spiky. rate() averages across the entire window and is better for alerting and smoother graphs.
Question 10
Which of the following is a valid PromQL aggregation operator?
A.mean
B.topk
C.median
D.rate
Correct answer: B
topk is a built-in aggregation operator returning the k largest elements. There is no mean or median operator (use avg and quantile), and rate is a function, not an aggregation operator.
Question 11
What does rate(http_requests_total[5m]) compute?
A.The total number of requests received in the last 5 minutes
B.The per-second average rate of increase of the counter over the last 5 minutes
C.The instantaneous current value of the counter
D.The maximum value observed during the 5-minute window
Correct answer: B
rate() returns the per-second average rate of increase of a counter over the range, automatically accounting for counter resets. For a raw total over the window you would use increase().
Question 12
The label matcher =~ performs what kind of match?
A.A regular-expression match on the label value
B.An exact string inequality test
C.A numeric greater-than comparison
D.A negated regular-expression match
Correct answer: A
=~ matches label values against a regular expression. The negated regex matcher is !~, while = and != perform exact equality and inequality.
Question 13
For many-to-one (or one-to-many) vector matching, which keywords are used?
A.by and without
B.and / or
C.group_left and group_right
D.offset and @
Correct answer: C
group_left and group_right specify the 'many' side in many-to-one and one-to-many matches, allowing labels from the one side to be carried over. by/without are for aggregation grouping.
Question 14
In PromQL, evaluating a bare metric name such as http_requests_total returns:
A.A single scalar number
B.A range vector covering the last minute
C.A string value
D.An instant vector — one sample per matching series at the evaluation time
Correct answer: D
A metric selector returns an instant vector: a set containing the single most recent sample for each matching series at the query's evaluation timestamp.
Instrumentation and Exporters (3 questions)
Question 15
When instrumenting latency, which metric type lets you later compute aggregatable percentiles across instances?
A.Counter
B.Gauge
C.Summary
D.Histogram
Correct answer: D
Histograms record observations into buckets that can be summed across instances and fed to histogram_quantile() server-side. Summary quantiles are computed per client and cannot be aggregated.
Question 16
What is the recommended way to expose metrics from a short-lived batch job?
A.Push the metrics to a Pushgateway, which Prometheus then scrapes
B.Have Prometheus scrape the job directly as it exits
C.Write the metrics to a local log file for later parsing
D.Email the metrics to Alertmanager
Correct answer: A
Because a batch job may finish before Prometheus can scrape it, it should push its final metrics to a Pushgateway. Prometheus then scrapes those persisted values from the gateway.
Question 17
What does the job label typically indicate in Prometheus?
A.The name of the server where metrics are stored
B.The role of a target or group of targets
C.The version of Prometheus running
D.The current alert status
Correct answer: B
The job label typically groups targets by their function or role in the infrastructure.
Alerting & Dashboarding (3 questions)
Question 18
Which component handles alerts generated by Prometheus?
A.Node Exporter
B.Alertmanager
C.Blackbox Exporter
D.Prometheus Server
Correct answer: B
Alertmanager handles alerts generated by Prometheus, managing notification routing and silencing.
Question 19
Annotations on an alert are typically used to:
A.Identify and route the alert through Alertmanager
B.Provide human-readable descriptive information such as a summary or runbook link
C.Define the numeric alerting threshold
D.Set the data retention period
Correct answer: B
Annotations carry descriptive, human-oriented content like a summary, a longer description, or a runbook URL. Routing and grouping decisions are driven by labels, not annotations.
Question 20
A recording rule is used to:
A.Send notifications directly to receivers
B.Silence noisy alerts during maintenance
C.Precompute frequently used or expensive expressions and save the result as a new series
D.Discover and register scrape targets
Correct answer: C
Recording rules evaluate an expression on a schedule and store the result as a new time series, speeding up dashboards and alerts that repeatedly use costly queries. They do not send notifications.
Score yourself
The Linux Foundation lists a passing score of 75% for PCA. On a set of 20 questions that is 15 correct. Reaching 15 here is a useful signal and not a prediction: 20 questions is a small sample, and the real exam covers every domain in its published weights. Use the table below to decide what to do next. Count only the questions you answered correctly without help, and be strict about it: a question you got right for the wrong reason belongs in the missed pile. Then repeat the exercise a few days later with a different set from the full bank so that you are measuring understanding and not memory of these specific questions.
Count your misses per domain, using the headings above. One missed question in a domain is noise; three misses in one domain is a study plan. The practice mode on GoldenKube does this sorting for you by sending you back to the domains where you are weakest.
Correct out of 20
What it suggests
What to do next
18 to 20
The basics are solid in the domains sampled.
Move to harder mixed practice and check the domains this set did not cover.
15 to 17
Close to the benchmark, with some gaps.
Review every missed explanation, then practise the domains where you missed questions.
11 to 14
Several domains need real study.
Read the official curriculum for those domains, then return to practice.
0 to 10
You are early in your preparation.
Start from the curriculum and the project documentation, then retry this set.
Observability Concepts: what it tests
Observability Concepts is the vocabulary domain of PCA. Expect the difference between monitoring and observability, the three signals of metrics, logs and traces, service level indicators, objectives and agreements, the pull and push models of collecting data, and ideas such as cardinality and the cost of high-volume series. Questions are usually definitions or small scenarios asking which signal answers a given question. A common trap is reaching for logs when the question asks for an aggregate trend, which is a metrics job, or for metrics when the question asks about one request's path, which is a tracing job. Short and conceptual, this domain is worth getting fully right.
Write down one question that each signal answers best, and one that it answers badly. If you can do that for all three, the concept questions stop being tricky.
Prometheus Fundamentals: what it tests
Prometheus Fundamentals covers how Prometheus is built and how data gets into it. Expect the architecture with the server, time series database, exporters, Pushgateway, Alertmanager and service discovery, the data model of a metric name plus labels, the four metric types (counter, gauge, histogram and summary), scrape configuration, relabelling, retention, and options such as federation and remote write. Questions ask where a behaviour belongs or what a configuration does. A typical trap is the metric type: using a gauge where a counter is correct changes which queries are meaningful, so recall the use case for each type before looking at the options.
Sketch the data flow once, from target to scrape to storage to query to alert, and label every arrow with the component that performs it. Most architecture questions then answer themselves.
PromQL: what it tests
PromQL is the largest PCA domain, and it rewards practice over reading. Expect instant and range vector selectors, label matchers, the functions rate, irate and increase, aggregation operators with by and without, histogram_quantile, offsets and subqueries, and binary operators with vector matching. Questions give a query or a goal and ask for the result or the right expression. The classic traps are applying rate to a gauge, forgetting that rate needs a range selector, aggregating before taking a rate instead of after, and misreading what a result drops when labels do not match. Reading the query out loud, one operator at a time, catches most of them.
Write ten queries by hand against a metric you know and predict the shape of each result before you run it. The prediction step is what builds the instinct that the exam tests.
Instrumentation and Exporters: what it tests
Instrumentation and Exporters covers how applications and systems expose metrics. Expect the official client libraries, the text exposition format, metric and label naming conventions, base units, when to use a counter, a gauge or a histogram in your own code, well-known exporters such as the node exporter, writing a custom exporter, and when the Pushgateway is appropriate. A common trap is the Pushgateway: it exists for short-lived batch jobs and is not a general way to send metrics from services. Another is high-cardinality labels such as user identifiers, which look convenient and quietly make a server expensive to run. Expect questions about naming, types and the choice of mechanism for a given source of metrics.
Take a service you know and list five metrics you would add, each with its type, name and labels. Checking those against the naming conventions is the best preparation for the naming questions.
Alerting & Dashboarding: what it tests
Alerting and Dashboarding covers turning data into notifications and views. Expect alerting rules and the for clause, the roles of Prometheus and Alertmanager, grouping, routing, inhibition and silences, notification receivers, alerting on symptoms rather than causes, and how dashboards are built from queries. Questions ask what a setting changes or where a behaviour is configured. The recurring trap is the split of responsibility: Prometheus evaluates rules and fires alerts, while Alertmanager decides what to do with them. Another is writing an alert that flaps because the for duration is missing, which the exam describes as noisy alerts. Questions in this area often hinge on knowing which of the two programs owns a given behaviour.
Trace one alert end to end: rule, pending, firing, grouped, routed, notified, silenced or inhibited. Naming the owner of each step settles most of the questions in this domain.
What a 20-question PCA sample can and cannot tell you
A set of 20 questions is a screening test, not a measurement. With only a few questions per domain, one lucky guess or one unlucky wording moves your result by five percentage points, and a domain with only two questions behind it says almost nothing about that domain's true strength. The set is also fixed, so after one read-through it measures your memory of these items and not your command of PCA. Use it to choose where to start studying, not to decide whether you are ready, and then rely on the larger bank, where questions repeat only after a gap and the app tracks which domains stay weak.
A second limit is style. Written practice questions share one author voice, while a real exam draws on many writers and reads differently. Treat any practice score, here or elsewhere, as one input next to the official curriculum, hands-on experience with the technology, and the documentation of the project itself.
A one-week plan after the PCA sample set
After the sample set, spend a week turning the result into a plan. Start with the domains where you missed questions and read the matching part of the official curriculum before you try more questions, because practice is far more useful once you can explain the topic. Return to practice in short daily sessions, review every missed question the next day, and leave the last day for a fresh mixed set. Keep a simple note of which concepts caused misses; the same three or four ideas usually explain most of them, and naming them is what turns practice into progress.
The largest PCA domain is PromQL, at 28% of the published weights. If two domains tie for weakest in your tally, start with the heavier one, and keep the other 4 domains ticking over with short reviews.
Adjust the pace to your calendar. If you have more time, spread the same steps over two or three weeks and add a second mixed session at the end of each week. If you have less, keep days 1 to 3 and day 6 and accept a rougher picture. What matters is that every study session starts from a specific weak domain and ends with a check, not that the plan is followed to the letter.
Day
Task
1
Tally misses per domain and rank the domains from weakest to strongest.
2
Read the official curriculum and documentation for the weakest domain.
3
Practise that domain in the app and read every explanation.
4
Repeat the exercise for the second weakest domain.
5
Review the questions you missed on days 3 and 4 without looking at notes.
6
Do a mixed session across all domains and compare the result with this page.
7
Rest, then retake the 20 questions on this page and note which misses persist.
Practice the full PCA bank
These 20 questions are a small slice of the GoldenKube PCA bank, which holds more than 500 questions. In practice mode the app serves the questions you are due to review first, then the domain where you are weakest, and it brings missed questions back after a gap until you answer them correctly. You can answer 10 questions as a guest without an account; after that a free account keeps your progress, and no credit card is asked for at any point. Each answer comes with the same kind of written explanation you saw on this page, so a wrong answer is always followed by the reasoning that fixes it.
Practice mode shows one question at a time and gives the verdict and explanation as soon as you answer. A focus option lets you stay inside a single domain until it feels solid, which is the natural next step after finding your weakest domain on this page. There is no timer and no exam simulation in practice mode; the goal is learning, not performance under pressure.
No. They are practice questions written for GoldenKube to cover the published PCA domains. GoldenKube does not have the real exam, does not claim any overlap with it, and does not sell or share exam content. Use them to find gaps in your understanding, not to predict what you will be asked.
How many PCA questions does GoldenKube have?
The PCA bank holds more than 500 questions as of 2026-10-23. This page shows 20 of them. The full bank is available in practice mode, where the app chooses what to show you next based on what you have already answered and what you got wrong.
Do the answers come with explanations?
Yes. Every question on this page has the correct answer marked and a written explanation of why it is correct and, where useful, why the other options are not. Only questions that have an explanation are published on public pages, and where a question carries references they are listed under the explanation.
Is this free?
Yes. Reading this page needs no account. In the practice app you can answer 10 questions as a guest, and then you create a free account to continue. No credit card is requested at any point.
How were these 20 questions chosen?
A script picks them with a fixed seed, so the set does not change between visits. It considers only questions that have a written explanation and tries to follow the published domain weights. The domain labels in the bank do not always match the official wording, so a domain can be thinly represented in a single set of 20; the full bank is the better measure of coverage.
How often is this page updated?
The exam facts and the domain list were last checked against the Linux Foundation pages on 2026-10-23, and the page shows its update date at the top. When the curriculum changes or questions are added, corrected or removed in the bank, the page is reviewed again.
Independent community project, not affiliated with or endorsed by CNCF or The Linux Foundation.