PCA Observability Concepts Practice Test 3 — 15 Questions

Question 1

You need to measure the P99 latency of your service. Which PromQL function do you use on a Histogram?

Accepted Answer

histogram_quantile(). histogram_quantile() is the correct PromQL function to compute quantiles (like P99) from a histogram metric. It takes a φ (phi) value between 0 and 1 and a histogram bucket series, then interpolates the quantile from the cumulative bucket counts. For P99, you would use histogram_quantile(0.99, ...). This function is specifically designed for Prometheus histograms, which expose _bucket time series with an le label.

Answer

avg()

Answer

sum()

Answer

rate()

Question 2

What is the primary function of the 'scrape_timeout' configuration?

Accepted Answer

It limits the duration of the HTTP connection.. It defines how long Prometheus waits for a response from the target before considering the scrape a failure.

Answer

It determines how long to keep metrics in memory.

Answer

It sets the interval between scrapes.

Answer

It restricts the size of the metric payload.

Question 3

How does the 'stale marker' work in Prometheus?

Accepted Answer

It prevents the graph from showing data after the target goes down.. When a target stops responding, Prometheus writes a stale marker to the TSDB so that graphs stop at the last known value rather than showing an infinite line or incorrect data.

Answer

It restarts the scrape job.

Answer

It sends an alert.

Answer

It deletes the metric data.

Question 4

If you have a metric `http_requests_total` and want the increase over 5 minutes, what is the best PromQL syntax?

Accepted Answer

increase(http_requests_total[5m]). The `increase()` function is specifically designed to calculate the increase of counter values over a specified range.

Answer

delta(http_requests_total[5m])

Answer

rate(http_requests_total[5m])

Answer

sum(http_requests_total[5m])

Question 5

What is the primary goal of observability?

Accepted Answer

To understand system state from external outputs.. Observability aims to answer 'why' a system is in its current state by understanding its internal state through outputs (metrics, logs, traces).

Answer

To store as many logs as possible.

Answer

To automate code deployments.

Answer

To replace testing.

Question 6

Why are 'logs' essential for debugging, even with perfect metrics?

Accepted Answer

Logs contain detailed context about specific events.. Logs provide the granular, detailed context (like specific error messages or stack traces) that metrics cannot provide.

Answer

Metrics are too slow.

Answer

Logs are free to store.

Answer

Metrics cannot be used for debugging.

Question 7

When evaluating the 'Traffic' golden signal, what is the best metric to watch for a web service?

Accepted Answer

Requests per second.. Request rate (requests per second) is the standard way to measure traffic volume.

Answer

CPU usage percentage.

Answer

HTTP error count.

Answer

Response latency.

Question 8

Which of these is NOT a Golden Signal?

Accepted Answer

Storage. The Four Golden Signals are Latency, Traffic, Errors, and Saturation. 'Storage' is not one of them.

Answer

Latency

Answer

Saturation

Answer

Errors

Question 9

What is the benefit of the 'service discovery' feature in Prometheus?

Accepted Answer

It simplifies management in dynamic environments.. Prometheus service discovery automatically detects and updates scrape targets from dynamic sources such as Kubernetes, EC2, Consul, or DNS. In environments where instances are constantly created and destroyed, this eliminates manual target configuration and keeps monitoring in sync with the actual infrastructure. The primary benefit is simplified management in dynamic environments.

Answer

It improves query performance.

Answer

It encrypts the data.

Answer

It reduces the storage size.

Question 10

Which THREE items are standard techniques to reduce metric cardinality?

Accepted Answer

Aggregating raw metrics into broader categories.. Option A is correct because aggregating raw metrics into broader categories (for example, rolling per-endpoint series up to a per-service or per-method level) collapses many distinct label combinations into fewer time series, directly lowering cardinality. Option C is correct because dropping unused labels via relabeling (e.g., Prometheus metric_relabel_configs with action: labeldrop) removes high-variation or unnecessary label dimensions before ingestion, which is a standard cardinality-reduction technique. Option D is correct because histograms with fewer buckets reduce the number of derived series per histogram, since each bucket becomes its own time series with a le label, so cutting buckets directly cuts cardinality. Option B does not belong because increasing the scrape interval only changes sampling frequency, not the number of distinct label sets, so cardinality is unchanged. Option E does not belong because adding unique IDs to every label does the opposite: it creates a unique series per ID and explodes cardinality.

Answer

Increasing the scrape interval.

Answer

Adding unique IDs to every label.

Question 11

Which TWO of these are potential side effects of high-cardinality metrics?

Accepted Answer

High memory consumption on the Prometheus server.. Memory exhaustion (OOM) and slow query performance are the direct results of high cardinality.

Answer

Reduced CPU usage on target.

Answer

Faster alert evaluation.

Answer

Decreased network bandwidth.

Question 12

Which THREE of these represent the Four Golden Signals?

Accepted Answer

Errors. The Four Golden Signals, as defined in Google's SRE guidance, are Latency, Traffic, Errors, and Saturation, so option E (Latency) is correct because it measures the time taken to serve requests, typically distinguishing successful-request latency from failed-request latency. Option C (Traffic) is correct because it quantifies demand on the system, such as HTTP requests per second or network I/O rate. Option B (Errors) is correct because it captures the rate of failed requests, including explicit failures like HTTP 5xx responses and implicit failures such as wrong content returned with a 200 status. The remaining options do not belong: Memory Usage (A) is only one possible component of saturation rather than a golden signal itself, and Availability (D) is a broader service-level objective concept, not one of the four signals.

Answer

Memory Usage

Answer

Availability

Question 13

Which TWO of the following are primary components of the Prometheus monitoring architecture?

Accepted Answer

Alertmanager. The Prometheus Server (D) is the core component of the Prometheus architecture: it scrapes metrics from instrumented targets over HTTP, stores them in its local time-series database (TSDB), and evaluates PromQL recording and alerting rules. Alertmanager (B) is the other primary component, receiving alerts fired by the Prometheus Server and handling deduplication, grouping, silencing, inhibition, and routing them to receivers such as email, Slack, or PagerDuty. Fluentd (A) is a log collection and forwarding tool, not part of Prometheus's metric-based monitoring pipeline. Zipkin (C) is a distributed tracing system, and Kibana (E) is a visualization front end for Elasticsearch logs, neither of which are Prometheus components.

Answer

Fluentd

Answer

Zipkin

Answer

Kibana

Question 14

Which TWO of the following are examples of metric data?

Accepted Answer

Current CPU usage percentage.. Metric data consists of numeric measurements captured over time, typically aggregated as counters, gauges, or histograms. Option B (Current CPU usage percentage) is correct because it is a numeric gauge value representing resource utilization at a point in time, exactly the kind of time-series measurement monitoring systems collect. Option D (Total number of completed HTTP requests) is correct because it is a numeric counter that monotonically increases as requests are served, a canonical metric used for rate and throughput calculations. Option A (a full database SQL query statement) is not metric data; it is a text/string artifact more suited to logs or traces. Option C (a line of text in an access log file) is log data, not a numeric measurement. Option E (the stack trace of an application error) is diagnostic trace/log data, not metric data.

Answer

A full database SQL query statement.

Answer

A line of text in an access log file.

Answer

The stack trace of an application error.

Question 15

Which TWO of these are true about the 'Pull' model?

Accepted Answer

Prometheus initiates the scrape request.. Option C is correct because in the Pull model Prometheus itself initiates the HTTP scrape request to each target's /metrics endpoint on a schedule, rather than waiting for targets to send data. Option D is correct because, since Prometheus controls the scrape cycle, it can evaluate the `up` metric after each scrape and mark a target as down when the scrape fails or times out, enabling alerting on target availability. Option A is not correct because the pull model's load is predictable and driven by Prometheus's configured scrape intervals and target list, not by unpredictable target behavior. Option B is not correct because targets only need to expose an HTTP metrics endpoint; a push-client is required in push-based models, not in the pull model. Option E is not correct because Prometheus can scrape remote targets over the network, not just local metrics.

Answer

The load on the server is unpredictable.

Answer

Targets must run a push-client.

Answer

It is limited to local metrics.