PCA Practice Test 9 — 10 Questions

Question 1

Which operator has higher precedence in PromQL: exponential (^), multiplication (*), or addition (+)?

Accepted Answer

Exponentiation (^). Exponentiation has the highest precedence in PromQL, followed by multiplication/division, and then addition/subtraction.

Answer

Addition (+)

Answer

All arithmetic operators share equal precedence.

Answer

Multiplication (*)

Question 2

An operator needs to find the rate of change of a counter using deriv(). Why is this considered a bad practice in Prometheus?

Accepted Answer

deriv() does not account for counter resets, causing massive negative drops when counters reset.. deriv() is designed for linear regression on gauge metrics over range vectors and does not handle counter resets, leading to incorrect downward spikes when counters reset.

Answer

deriv() automatically multiplies output by an undocumented scaling factor.

Answer

deriv() requires native histograms, which are incompatible with standard counters.

Answer

deriv() only works with scalar values and rejects time series vectors.

Question 3

What is the primary operational difference between the 'rate()' and 'irate()' functions in PromQL?

Accepted Answer

rate() averages the per-second rate over the entire specified range vector, while irate() calculates the rate based on the last two data points in the range vector.. rate() calculates the average per-second rate over the whole range vector, smoothing spikes, whereas irate() calculates the instantaneous rate using only the last two points.

Answer

rate() can be used with instant vectors, whereas irate() requires range vectors.

Answer

irate() automatically accounts for counter resets, whereas rate() does not.

Answer

irate() supports gauges, while rate() only supports counters.

Question 4

You are using the 'predict_linear()' function to forecast disk space exhaustion based on the metric 'node_filesystem_free_bytes'. What type of input data does predict_linear require?

Accepted Answer

A range vector of gauge values and a time duration scalar.. predict_linear() requires a range vector of gauge values (e.g., node_filesystem_free_bytes[1h]) and a scalar representing the number of seconds into the future to predict. It uses linear regression over the range to forecast the value at the future time.

Answer

An instant vector representing current disk space.

Answer

Two instant vectors joined via an 'on' clause.

Answer

A counter range vector wrapped in a rate() function.

Question 5

Which PromQL function is specifically designed to calculate the per-second average rate of increase of a time series over a given range vector?

Accepted Answer

rate(). The rate() function in PromQL is specifically designed to calculate the per-second average rate of increase of a time series over a given range vector. It takes a range vector (e.g., http_requests_total[5m]) and returns the average per-second growth rate, handling counter resets automatically. This is the canonical function for computing rates of counters.

Answer

deriv()

Answer

irate()

Answer

increase()

Question 6

You want to find the top 3 jobs with the highest CPU usage using the metric 'process_cpu_seconds_total'. Which PromQL query structure is correct?

Accepted Answer

topk(3, rate(process_cpu_seconds_total[5m])). The correct PromQL structure is topk(3, rate(process_cpu_seconds_total[5m])). topk takes the instant-vector result of rate() over a 5-minute range vector and returns the 3 series with the highest values. The rate() function is required because process_cpu_seconds_total is a counter, and raw counter values would rank by total lifetime CPU rather than current usage.

Answer

bottomk(3, rate(process_cpu_seconds_total[5m]))

Answer

topk(3, process_cpu_seconds_total[5m])

Answer

topk(rate(process_cpu_seconds_total[5m]), 3)

Question 7

You want to notify different teams based on the 'service' label of an alert. Where should you configure this logic?

Accepted Answer

In the Alertmanager configuration. The Alertmanager configuration file contains the routing tree that maps alerts to receivers based on labels.

Answer

In the node exporter config

Answer

In the Grafana data source

Answer

In the Prometheus alerting rule

Question 8

You have a recording rule that references a metric that doesn't exist. What is the impact?

Accepted Answer

The rule evaluation results in no data. The recording rule will be evaluated, but because the expression returns no data, no new time series will be created for that rule.

Answer

Prometheus logs a critical error

Answer

The entire server crashes

Answer

The rule is automatically disabled

Question 9

You need to track the number of requests received by your application. Which metric type should you choose?

Accepted Answer

Counter. A counter is a cumulative metric that represents a single monotonically increasing counter whose value can only increase or be reset to zero on restart.

Answer

Summary

Answer

Gauge

Answer

Histogram

Question 10

A developer asks why they should use Distributed Traces instead of simply using high-cardinality metrics to debug a slow request. What is the primary advantage of traces?

Accepted Answer

Traces provide end-to-end request context across microservices.. Distributed traces capture the full causal chain of a request as it flows through multiple microservices, including parent-child span relationships, timing, and metadata. This end-to-end context lets you pinpoint which service or span in a call graph introduced latency, which high-cardinality metrics alone cannot do because metrics aggregate and lose per-request causality.

Answer

Traces are natively generated by Prometheus.

Answer

Traces have lower storage requirements.

Answer

Metrics cannot be used for alerting.