I85PREP
⌘ K
Harness Certified Chaos Engineering study materials
Harness Certified Chaos Engineering study resources

Harness Certified Chaos Engineering practice questions

One question at a time. Build your Harness Certified Chaos Engineering understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Chaos Engineering Fundamentals

In chaos engineering, what is the primary purpose of stating a steady-state hypothesis before running an experiment?

  1. To define a measurable baseline of normal behavior that the experiment will attempt to falsify Correct answer.
  2. To record the postmortem actions assigned after an incident is resolved
  3. To specify which team owns the service being tested
  4. To document the exact sequence of fault injections the tooling will execute

Explanation

A steady-state hypothesis expresses what normal looks like in measurable terms so the experiment can try to disprove it. Fault injection details and ownership are separate concerns; the hypothesis is about observable system behavior.

Question 2Chaos Engineering Fundamentals

Which statement best captures the core principle behind chaos engineering as a discipline?

  1. It is a testing phase performed only once before the first production release
  2. It guarantees that production incidents will never occur once experiments pass
  3. It builds confidence in a system by deliberately experimenting on it to discover weaknesses Correct answer.
  4. It replaces monitoring by generating synthetic traffic during peak hours

Explanation

Chaos engineering is an experimental discipline that surfaces unknown weaknesses before they cause outages, thereby building confidence. It complements monitoring and testing rather than replacing them, and it never guarantees the absence of incidents.

Question 3Chaos Engineering Fundamentals

A team wants the smallest possible blast radius for a first experiment on a payment service. Which approach best fits that goal?

  1. Inject a fault into every production instance during business hours
  2. Randomly terminate pods across all clusters simultaneously
  3. Run the fault in a staging environment with production-like traffic Correct answer.
  4. Disable the entire region's network for ten minutes

Explanation

Blast radius is the scope of impact an experiment can cause, so starting small and controlled is the guiding rule. A staging environment with realistic traffic limits exposure while still yielding useful signal, whereas the other options maximize impact.

Question 4Chaos Engineering Fundamentals

What does the abort condition (sometimes called the safety check) in a chaos experiment provide?

  1. A threshold that, when breached, stops the experiment to limit harm Correct answer.
  2. A list of services that are exempt from all future experiments
  3. The schedule for rotating on-call engineers
  4. The maximum number of experiments allowed per quarter

Explanation

The abort condition is a predefined guardrail: if metrics breach a threshold during the experiment, the fault is halted automatically. This keeps the experiment from causing unacceptable damage and is distinct from scheduling or exemption lists.

Question 5Chaos Engineering Fundamentals

Which of the following is the clearest example of a chaos experiment's independent variable?

  1. The latency threshold defined in the hypothesis
  2. The measured error rate of the checkout service
  3. The injected network delay of 200 milliseconds Correct answer.
  4. The number of users in the steady-state baseline

Explanation

The independent variable is the condition the experimenter deliberately manipulates, such as an injected delay. Error rate and user counts are observed outcomes, and the threshold is part of the hypothesis rather than the manipulated input.

Question 6Chaos Engineering Fundamentals

A team runs a chaos experiment and observes that the system behaves as the hypothesis predicted. What is the most appropriate conclusion?

  1. Confidence in the system under that condition increased, and the next experiment can probe further Correct answer.
  2. The fault injection tooling must be misconfigured
  3. The system has no remaining weaknesses and experiments can stop
  4. The experiment failed because nothing broke

Explanation

A passing experiment is a success: it increases confidence in the system's resilience under the tested condition. It does not prove the absence of all weaknesses, so the practice continues with new hypotheses and conditions.

Question 7Chaos Engineering Fundamentals

Why do mature chaos programs prefer to run experiments in production rather than only in pre-production environments?

  1. Production experiments never require an abort condition
  2. Pre-production environments cannot run fault injection tools at all
  3. Running in production removes the need for observability
  4. Production is the only place where real dependencies, traffic patterns, and configuration exist Correct answer.

Explanation

Production captures real traffic, third-party dependencies, and configuration drift that staging cannot fully replicate, so it reveals weaknesses that only manifest there. This does not remove the need for safety controls or observability; those become more important.

Question 8Chaos Engineering Fundamentals

Which practice best supports the chaos engineering principle of minimizing the impact of experiments on real customers?

  1. Running every experiment at the same time to finish quickly
  2. Skipping the steady-state hypothesis to save time
  3. Starting with small-scope faults and expanding only after confidence grows Correct answer.
  4. Removing all monitoring to avoid alert noise during the experiment

Explanation

Gradually increasing scope lets teams validate safety controls before exposing more of the system, which limits customer impact. Removing monitoring, batching all experiments, or skipping the hypothesis all increase risk rather than contain it.

Question 9Chaos Engineering Fundamentals

A chaos experiment injects a 5% packet loss on one service while the team watches a dashboard. Which element makes this a controlled experiment rather than random disruption?

  1. The fault is applied only when an outage is already underway
  2. The fault is applied without any measurement
  3. The fault is applied to a scoped target with defined magnitude and observation Correct answer.
  4. The fault is applied to the whole datacenter

Explanation

Controlled experimentation requires a scoped target, a known magnitude, and observation of the outcome. Unmeasured, unbounded, or incident-triggered disruption lacks the deliberate structure that distinguishes chaos engineering from random breakage.

Question 10Chaos Engineering Fundamentals

What role does observability play in a chaos experiment?

  1. It replaces the need for a hypothesis by auto-detecting faults
  2. It prevents the fault from being injected in the first place
  3. It guarantees the experiment will pass
  4. It provides the metrics and traces needed to judge whether steady state held Correct answer.

Explanation

Observability supplies the signals that let the team compare actual behavior against the steady-state hypothesis. It does not substitute for the hypothesis, block injection, or determine the outcome in advance.

Question 11Chaos Engineering Fundamentals

A team defines a hypothesis that a service stays available when one of three replicas is terminated. Which experiment most directly tests this hypothesis?

  1. Terminating one replica and measuring availability against the threshold Correct answer.
  2. Rotating the database credentials
  3. Increasing the number of replicas to ten permanently
  4. Scaling the cluster down to zero replicas

Explanation

The experiment must match the hypothesis: terminating exactly one replica tests the stated tolerance. Scaling to zero or unrelated changes do not test the specific claim about surviving a single replica loss.

Question 12Chaos Engineering Fundamentals

Which outcome is the strongest signal that a chaos experiment found a genuine weakness?

  1. A monitored service breached its steady-state threshold after the fault was injected Correct answer.
  2. The experiment ran for the planned duration
  3. The dashboard showed normal traffic throughout
  4. The on-call engineer was paged for an unrelated alert

Explanation

A weakness is revealed when injected conditions cause the system to deviate from steady state in a monitored way. Normal traffic, duration, or unrelated pages do not demonstrate a causal resilience gap.

Question 13Chaos Engineering Fundamentals

Why should chaos experiments be automated and run continuously in a mature practice?

  1. To eliminate the need for any human review of results
  2. To avoid ever writing new hypotheses
  3. To ensure resilience is repeatedly validated as the system and its dependencies change Correct answer.
  4. To reduce the number of metrics collected

Explanation

Systems evolve constantly, so resilience must be re-validated continuously rather than proven once. Automation supports repetition, but humans still review results and hypotheses; it does not reduce observability needs.

Question 14Chaos Engineering Fundamentals

A team wants to test how a service behaves when a downstream dependency becomes slow. Which injection best models this scenario?

  1. Adding latency to calls to the downstream dependency Correct answer.
  2. Terminating the service's own pod
  3. Deleting the service's deployment manifest
  4. Increasing the service's CPU limit

Explanation

Modeling a slow dependency means introducing delay on the calls to that dependency so timeouts and fallbacks are exercised. Terminating the caller, deleting manifests, or raising CPU limits do not reproduce downstream slowness.

Question 15Chaos Engineering Fundamentals

Which is the best reason to define a clear start time and end time for a chaos experiment?

  1. To avoid recording the results
  2. To make the experiment impossible to abort
  3. To guarantee the system will fail
  4. To ensure the fault is bounded and any lingering effects are removed afterward Correct answer.

Explanation

Bounding the experiment in time keeps the fault scoped and ensures cleanup so side effects do not persist. It does not prevent aborting, guarantee failure, or remove the need to record results.

Question 16Chaos Engineering Fundamentals

A team's steady-state hypothesis relies only on average request latency. Why is this a weak hypothesis?

  1. Averages are always higher than any individual request
  2. Error rates are the only valid steady-state metric
  3. Latency cannot be measured in production
  4. Averages hide tail latency and error distribution that users actually experience Correct answer.

Explanation

Averages can mask severe tail latency and error spikes affecting a subset of users, so the hypothesis may miss real degradation. Percentiles and error rates better capture user-impacting behavior than a single mean.

Question 17Chaos Engineering Fundamentals

Which scenario best illustrates the difference between chaos engineering and conventional functional testing?

  1. Functional testing injects random faults to find unknown behavior
  2. Both disciplines only verify documented requirements
  3. Chaos engineering verifies only unit-level code paths
  4. Functional testing verifies known expected outputs; chaos engineering probes unknown failure behavior under injected conditions Correct answer.

Explanation

Functional testing checks that the system meets known specifications, while chaos engineering explores how the system behaves under adverse, often unanticipated conditions. The other options misattribute the exploratory fault-injection role to functional testing.

Question 18Chaos Engineering Fundamentals

What is the purpose of running a chaos experiment first in a controlled, low-traffic window?

  1. To reduce the number of users potentially affected while validating the experiment design Correct answer.
  2. To increase blast radius gradually without measurement
  3. To remove the need for abort conditions
  4. To ensure the fault cannot be observed

Explanation

A low-traffic window limits exposure while the team validates the experiment design and safety controls. It does not hide the fault, remove abort conditions, or justify unmeasured expansion.

Question 19Chaos Engineering Fundamentals

A team observes that during a database failover experiment, the application kept serving requests but with elevated error rates below the abort threshold. How should this be interpreted?

  1. The experiment was invalid because errors occurred
  2. The system fully met the hypothesis with no degradation
  3. The abort threshold should be removed
  4. The system tolerated the failover but showed partial degradation worth investigating Correct answer.

Explanation

Staying within the abort threshold means the experiment remained safe, but elevated errors indicate partial degradation that merits investigation. This is a nuanced result, not a clean pass or a failed experiment.

Question 20Chaos Engineering Fundamentals

Which element of a chaos experiment is most directly responsible for preventing an experiment from causing a widespread outage?

  1. The number of dashboards created
  2. The length of the postmortem document
  3. The combination of a scoped blast radius and an abort condition Correct answer.
  4. The naming convention of the experiment files

Explanation

Safety comes from limiting the scope of the fault and halting automatically when thresholds are breached. Naming, dashboard counts, and document length do not constrain the experiment's impact.

Question 21Chaos Engineering Fundamentals

What does it mean to say a chaos experiment is 'falsifiable'?

  1. It requires no measurement to evaluate
  2. It cannot be repeated once executed
  3. It must always be proven true before running
  4. It is designed so that observations could show the hypothesis to be false Correct answer.

Explanation

A falsifiable hypothesis is stated so that concrete observations could contradict it, which is what makes the experiment meaningful. It does not presume the result, forbid repetition, or skip measurement.

Question 22Chaos Engineering Fundamentals

A team plans an experiment that injects a CPU stress fault on one node hosting a critical service. Which prerequisite most needs to be confirmed before injection?

  1. That the abort condition is disabled
  2. That all monitoring alerts are muted
  3. That the service has redundancy or failover so a single node fault is survivable Correct answer.
  4. That the node has no backups

Explanation

Before injecting a node-level fault, the team should confirm the service can tolerate losing that node, which bounds the expected impact. Muting alerts, disabling abort conditions, or lacking backups would undermine safety and observability.

Question 23Chaos Engineering Fundamentals

Which statement about the relationship between chaos engineering and incident response is most accurate?

  1. Chaos experiments should only run after incidents are resolved
  2. Incident response replaces the need for experiments
  3. Chaos engineering makes incident response unnecessary
  4. Chaos experiments can validate that runbooks and alerting work as intended Correct answer.

Explanation

Experiments can exercise detection and response paths, confirming that alerts fire and runbooks are usable. They complement incident response rather than replacing it or being confined to post-incident periods.

Question 24Chaos Engineering Fundamentals

A team wants to measure whether their service degrades gracefully when a dependency returns errors. Which hypothesis is best formed?

  1. The service will ignore all dependency responses permanently
  2. The service will crash immediately on any dependency error
  3. The service will return errors to users whenever any dependency fails
  4. The service will serve cached or fallback responses and keep error rate under a defined threshold Correct answer.

Explanation

A graceful-degradation hypothesis predicts a measurable fallback behavior with a bounded error rate. Predicting crashes, unbounded user errors, or permanent ignoring of dependencies does not describe graceful degradation.

Question 25Chaos Engineering Fundamentals

Why is it important to record the context and configuration of a chaos experiment alongside its results?

  1. So results can be reproduced and compared as the system evolves Correct answer.
  2. So the experiment can never be repeated
  3. To hide the fault injection details from the team
  4. To avoid documenting the hypothesis

Explanation

Recording context and configuration enables reproducibility and meaningful comparison over time as the system changes. It supports, rather than replaces, the hypothesis and keeps details transparent to the team.

Question 26Chaos Engineering Fundamentals

In chaos engineering, what does a steady-state hypothesis describe?

  1. A measurable range of normal behavior the system should keep during an experiment Correct answer.
  2. The exact fault that the experiment will inject into the target service
  3. The deployment pipeline stage where faults are first introduced
  4. The rollback procedure used if an experiment harms real customers

Explanation

A steady-state hypothesis defines observable, measurable normal behavior, such as latency or error rate bounds, that the system must preserve while a fault is active. It is expressed in terms of business or service metrics so the experiment can confirm or refute it rather than simply observing chaos.

Question 27Chaos Engineering Fundamentals

Which characteristic distinguishes chaos engineering from random fault injection?

  1. It relies on production traffic to expose faults
  2. It requires a dedicated staging environment
  3. It injects faults without any prior planning
  4. It follows a structured, hypothesis-driven experimental method Correct answer.

Explanation

Chaos engineering is a disciplined experimental practice: define a hypothesis, inject a controlled fault, measure, and learn. Random fault injection lacks the hypothesis and measurement discipline that make results interpretable and actionable.

Question 28Chaos Engineering Fundamentals

What is the primary purpose of defining a blast radius before running a chaos experiment?

  1. To estimate the cost of cloud resources used
  2. To decide which engineer leads the experiment
  3. To limit how much of the system and how many customers the fault can affect Correct answer.
  4. To set the duration of the observation window

Explanation

Blast radius scoping constrains the experiment's reach, for example to one instance, one zone, or a small percentage of traffic, so a surprising result does not become a widespread outage. It is a core safety control in any chaos program.

Question 29Chaos Engineering Fundamentals

A team wants to verify that their service remains healthy when a single availability zone becomes unreachable. Which experiment design element most directly models this?

  1. Rolling back the most recent deployment
  2. Terminating or isolating resources in one zone Correct answer.
  3. Increasing request rate on the load balancer
  4. Scaling the database read replicas upward

Explanation

Isolating or terminating resources in a single zone reproduces the zone-loss condition the team wants to test. The hypothesis then checks whether the service keeps serving within its steady-state bounds.

Question 30Chaos Engineering Fundamentals

Which metric type is most useful for detecting customer-visible harm during a chaos experiment?

  1. CPU utilization of the chaos controller
  2. Successful business transactions per minute Correct answer.
  3. Age of the container image
  4. Number of experiment runs completed

Explanation

Business-level signals such as completed transactions per minute directly reflect customer impact, which is what the steady-state hypothesis should protect. Infrastructure metrics can inform diagnosis but do not by themselves show user harm.

Question 31Chaos Engineering Fundamentals

Why should a chaos experiment include a defined abort condition?

  1. To prevent engineers from observing results
  2. To stop the experiment automatically when impact exceeds acceptable bounds Correct answer.
  3. To remove the need for a steady-state hypothesis
  4. To guarantee the experiment always produces a failure

Explanation

Abort conditions are safety thresholds, such as error rate or latency limits, that halt the fault when the impact grows beyond what was approved. They keep experiments from escalating into real incidents.

Question 32Chaos Engineering Fundamentals

What does the term 'game day' typically refer to in a chaos engineering program?

  1. A daily standup for the chaos tooling team
  2. A scheduled exercise that rehearses failure scenarios with the team Correct answer.
  3. A vendor demonstration of chaos tooling
  4. A code review of fault injection scripts

Explanation

A game day is a planned, often team-wide exercise where failure scenarios are injected and responses rehearsed in a controlled setting. It combines technical fault injection with human process validation.

Question 33Chaos Engineering Fundamentals

Which statement best captures the relationship between chaos engineering and resilience?

  1. Chaos engineering only applies to stateless services
  2. Chaos engineering is a method for validating and improving resilience Correct answer.
  3. Chaos engineering proves a system can never fail
  4. Chaos engineering replaces the need for resilience engineering

Explanation

Chaos engineering is an empirical method to test whether a system withstands turbulent conditions, and the findings drive resilience improvements. It validates resilience rather than guaranteeing the absence of failure.

Question 34Chaos Engineering Fundamentals

A team's hypothesis states that checkout success rate stays above 99% while one of four payment workers is killed. During the experiment, success rate drops to 97% for two minutes. What is the most accurate interpretation?

  1. The abort condition was necessarily triggered
  2. The steady state was maintained throughout
  3. The hypothesis was falsified, revealing a resilience gap Correct answer.
  4. The experiment is invalid because a worker was killed

Explanation

The observed success rate fell below the stated 99% bound, so the hypothesis was falsified and a genuine weakness was exposed. The team should investigate why losing one of four workers degraded checkout before deciding on remediation.

Question 35Chaos Engineering Fundamentals

Which experiment is the most appropriate first step for a team new to chaos engineering?

  1. Injecting a brief, reversible fault on a non-critical service in a controlled window Correct answer.
  2. Disabling all health checks in production indefinitely
  3. Simulating loss of an entire production region during peak hours
  4. Randomly killing pods across every cluster simultaneously

Explanation

New teams should start small with reversible, low-blast-radius faults in a controlled setting to build confidence and tooling maturity. Scope and impact grow as the team demonstrates competence and safety controls.

Question 36Chaos Engineering Fundamentals

A team notices that during a latency injection on a downstream API, their service's retries amplify load on that API. Which finding does this represent?

  1. A tooling misconfiguration with no system implication
  2. Proof that latency injection is unsafe in all cases
  3. A retry storm that the hypothesis should have bounded Correct answer.
  4. A successful demonstration of graceful degradation

Explanation

Retries multiplying under slow dependencies create a retry storm, an emergent failure mode that steady-state metrics like downstream request rate can reveal. The team should add bounded retries, backoff, or circuit breaking and re-test.

Question 37Chaos Engineering Fundamentals

Which practice best ensures chaos experiments are reproducible?

  1. Running experiments only when engineers remember to
  2. Recording the fault parameters, target scope, timing, and system version Correct answer.
  3. Varying fault parameters randomly on every run
  4. Keeping experiment definitions in individual engineers' notes

Explanation

Reproducibility requires capturing the exact configuration: fault type and magnitude, target selection, duration, and the system version under test. Stored, versioned experiment definitions let others rerun and compare results.

Question 38Chaos Engineering Fundamentals

A hypothesis predicts that p99 latency stays under 300 ms when a cache node is drained. Which measurement approach best tests it?

  1. Asking users whether the site felt slow
  2. Checking the average latency once after the experiment ends
  3. Reviewing CPU graphs from the previous week
  4. Sampling p99 latency continuously throughout the drain and recovery Correct answer.

Explanation

The hypothesis names a specific percentile and threshold, so the experiment must measure that percentile continuously across injection and recovery. Continuous sampling captures transient violations that a single end-of-run check would miss.

Question 39Chaos Engineering Fundamentals

Why is it valuable to run chaos experiments against dependencies the team does not control?

  1. To eliminate the need for timeouts and fallbacks
  2. To force the vendor to improve their reliability
  3. To confirm the vendor's SLA guarantees are accurate
  4. To test how the system behaves when third-party services fail or slow down Correct answer.

Explanation

Third-party dependencies will fail in ways outside the team's control, so experiments should simulate their failure or slowness to validate fallbacks, timeouts, and degradation paths. The goal is understanding local resilience, not auditing the vendor.

Question 40Chaos Engineering Fundamentals

Which element most distinguishes a chaos experiment's control group from its treatment run?

  1. The control run applies no fault while the treatment applies the planned fault Correct answer.
  2. The control run uses production data while the treatment uses synthetic data
  3. The control run is executed by a different team
  4. The control run lasts longer than the treatment

Explanation

A controlled experiment compares baseline behavior without the fault against behavior with the fault applied. Holding other conditions constant lets the team attribute observed differences to the injected fault.

Question 41Chaos Engineering Fundamentals

A team wants to confirm that their autoscaling policy recovers capacity after a sudden traffic surge. Which fault injection best models the trigger?

  1. Injecting disk I/O errors on the database
  2. Blocking DNS resolution for internal services
  3. Corrupting messages in the queue
  4. Generating a sharp, sustained increase in request volume Correct answer.

Explanation

Autoscaling responds to load, so a sharp and sustained traffic increase is the fault that exercises scale-out behavior. The hypothesis checks whether capacity and latency return to steady state within an expected time.

Question 42Chaos Engineering Fundamentals

What is the main reason to include the human on-call response in a chaos experiment's scope?

  1. To reduce the number of faults that need injection
  2. To avoid documenting the experiment
  3. To ensure the experiment never affects customers
  4. To measure whether alerts, runbooks, and escalation actually work Correct answer.

Explanation

Systems include people and processes, so experiments can validate that alerts fire, runbooks are accurate, and escalation paths work under realistic conditions. This surfaces operational gaps that metric-only experiments miss.

Question 43Chaos Engineering Fundamentals

Which condition makes an experiment's result inconclusive rather than a confirmed pass?

  1. The experiment ran for the planned duration
  2. The fault was injected but the key metric was not collected during the window Correct answer.
  3. The abort condition was never reached
  4. The steady-state metric stayed within bounds throughout

Explanation

If the metric that the hypothesis depends on was not captured while the fault was active, there is no evidence to confirm or refute the hypothesis. The run is inconclusive and should be repeated with proper instrumentation.

Question 44Chaos Engineering Fundamentals

A team plans to inject network latency between two services. Which safety prerequisite is most important before execution?

  1. Confirming the chaos tooling license tier
  2. Removing all dashboards to reduce noise
  3. Scheduling the experiment for a holiday weekend
  4. Ensuring monitoring, alerts, and a tested abort mechanism are active Correct answer.

Explanation

Before injecting any fault, the team must have observability in place to detect impact and a tested abort mechanism to stop the experiment quickly. These safeguards are prerequisites, not optional extras.

Question 45Chaos Engineering Fundamentals

Which statement best explains why chaos experiments should be tied to service-level objectives (SLOs)?

  1. SLOs define which engineers own the chaos tooling
  2. SLOs provide agreed thresholds for judging whether impact is acceptable Correct answer.
  3. SLOs guarantee the experiment will not affect users
  4. SLOs eliminate the need for a steady-state hypothesis

Explanation

SLOs encode the reliability targets the organization has committed to, so they give the experiment objective thresholds for acceptable behavior. A result that breaches an SLO is a clear, business-relevant finding.

Question 46Chaos Engineering Fundamentals

A team runs the same experiment weekly and it passes every time. What is the most reasonable next step for the program?

  1. Remove the steady-state hypothesis to simplify runs
  2. Increase fault severity or explore new failure modes within safety limits Correct answer.
  3. Stop all chaos testing because the system is proven resilient
  4. Run the experiment continuously without any abort condition

Explanation

Repeated passes indicate the current fault is well handled, so the program should evolve by increasing severity or targeting untested failure modes. This keeps experiments informative while staying within approved safety limits.

Question 47Chaos Engineering Fundamentals

An experiment injects packet loss on a service mesh sidecar. The service's error rate stays flat, but its p99 latency rises sharply and stays elevated after the fault ends. Which conclusion is best supported?

  1. The experiment had no effect and can be ignored
  2. The service tolerated errors but has a recovery problem worth investigating Correct answer.
  3. Packet loss cannot affect latency
  4. The steady-state hypothesis was fully satisfied

Explanation

Stable error rates suggest the service masked the loss, but persistent elevated latency after the fault ended points to a recovery defect such as backlog or connection pool exhaustion. The hypothesis should cover recovery, not only steady-state error rate.

Question 48Chaos Engineering Fundamentals

Which scenario best demonstrates that an experiment's blast radius was insufficiently controlled?

  1. The abort condition was defined but never triggered
  2. The fault targeted one pod but a shared dependency failed for all tenants Correct answer.
  3. The experiment ended at the scheduled time
  4. The fault was scoped to 1% of traffic and impact stayed within that group

Explanation

Blast radius control must account for shared dependencies, since a fault aimed at one pod can propagate to every tenant relying on that dependency. Effective scoping isolates the fault's effects to the intended population.

Question 49Chaos Engineering Fundamentals

A hypothesis claims that the system maintains steady state when a primary database fails over, but the team's only success metric is average response time. Why is this hypothesis weak?

  1. Averages are only valid in staging environments
  2. Failover experiments should never measure latency
  3. Average latency hides tail behavior and error rates that failover typically disturbs Correct answer.
  4. Average latency is too expensive to compute during experiments

Explanation

Failover often causes brief error spikes and tail latency excursions that averages smooth away, so a mean-only metric can mask real customer harm. A stronger hypothesis includes error rate and high percentiles alongside the mean.

Question 50Chaos Engineering Fundamentals

Two teams debate whether an experiment that caused a brief customer-visible error spike was successful. Which framing is most consistent with chaos engineering principles?

  1. It succeeded if the impact stayed within pre-approved limits and produced actionable learning Correct answer.
  2. It failed because the fault should have been injected without warning
  3. It succeeded only if no metrics were recorded
  4. It failed because any customer impact invalidates an experiment

Explanation

Chaos engineering accepts bounded, pre-approved impact in exchange for validated learning about system behavior. If the spike stayed within agreed limits and revealed something actionable, the experiment met its purpose.