Google Associate Cloud Engineer Practice Test 60
Question 590
A latency alert only checks mean Cloud Run request time and misses a serious degradation affecting the slowest users. Which signal should be added?
- Trace the end-to-end request through gateway and backend spans using correlated request identifiers.
- Set an unrelated CPU threshold without selecting the actual signal or metric labels.
- Increase Cloud Logging retention as a replacement for the required alert or metric definition.
- Monitor tail-latency percentiles such as P95 or P99 and error rates for representative request cohorts.
Explanation
Correct Answer
D:Mean response latency can hide severe outliers and cannot prove experience for slow users.
Incorrect Answers
A:It does not measure or notify on the signal requested by the case. It leaves the requested latency measurement unaddressed.
B:It does not measure or notify on the signal requested by the case. The specific requirement concerns the requested latency measurement.
C:Cloud Logging records telemetry, rather than directly enforcing the requested runtime behavior. The necessary distinction is the requested latency measurement.
Question 591
Two applications emit the same custom Cloud Monitoring latency metric. One exporter sends milliseconds while another sends seconds but both use the same metric descriptor. Graphs differ by orders of magnitude. What must be standardized?
- Increase time-series retention; retention automatically converts seconds to milliseconds for every stored sample.
- Use one documented metric unit and perform explicit unit conversion at the exporters; verify monitored-resource and metric labels.
- Give both exporters the same IAM group name; identity group membership controls the unit of an existing metric.
- Enable a Cloud Billing label so values are scaled to the price of the underlying VM.
Explanation
Correct Answer
B:Use one documented metric unit and perform explicit unit conversion at the exporters; verify monitored-resource and metric labels. All contributors to one custom metric must agree on value units and labels. Normalizing before ingestion makes combined time series comparable.
Incorrect Answers
A:Increase time-series retention; retention automatically converts seconds to milliseconds for every stored sample. Retention affects how long data is available, not the units or numerical meaning of samples already written.
C:Give both exporters the same IAM group name; identity group membership controls the unit of an existing metric. IAM controls permission to write metric data, not the scaling or unit conversion of reported numeric values.
D:Enable a Cloud Billing label so values are scaled to the price of the underlying VM. Billing labels describe cost attribution and cannot fix an inconsistent latency unit in a Monitoring time-series descriptor.
Question 592
An alert policy fires for every pod restart and overwhelms on-call staff during scheduled rolling upgrades. What should be tuned?
- Update and validate the alert policy's notification channels and incident escalation routing using an approved test.
- Apply a meaningful restart-rate threshold, duration and approved rollout context to distinguish unexpected failures from maintenance.
- Rename the monitored VM and assume metric-series selection is automatically corrected.
- Set an unrelated CPU threshold without selecting the actual signal or metric labels.
Explanation
Correct Answer
B:An isolated restart event does not necessarily indicate a service-level incident; policy conditions need operational context.
Incorrect Answers
A:It does not measure or notify on the signal requested by the case. The necessary distinction is the relevant monitoring signal or alert.
C:Renaming a resource does not alter the actual effective configuration. The case instead depends on the relevant monitoring signal or alert.
D:It does not measure or notify on the signal requested by the case. It leaves the relevant monitoring signal or alert unaddressed.
Question 593
A GKE Horizontal Pod Autoscaler reads a custom metric but its adapter returns a time series from namespace test instead of the production Pod set. The application metric exists. Which mapping should be corrected?
- Increase every Pod CPU request; CPU quantities automatically rewrite the custom metric adapter's namespace labels.
- Disable Cloud Logging; logging collection overrides the namespace label used by HPA custom metrics.
- Match the custom-metric source labels, namespace, workload selectors and adapter query to the intended production Pods.
- Grant public Cloud Storage object access; bucket visibility controls Kubernetes custom metric namespace resolution.
Explanation
Correct Answer
C:Match the custom-metric source labels, namespace, workload selectors and adapter query to the intended production Pods. Autoscaling uses the metric series mapped to the selected workload. Consistent namespace and selector dimensions are needed to evaluate the production workload.
Incorrect Answers
A:Increase every Pod CPU request; CPU quantities automatically rewrite the custom metric adapter's namespace labels. Resource requests affect scheduling and CPU-related autoscaling, not the namespace mapping for a separately defined custom metric.
B:Disable Cloud Logging; logging collection overrides the namespace label used by HPA custom metrics. Cloud Logging telemetry storage does not change the identity of custom metric time series returned by a Kubernetes metrics adapter.
D:Grant public Cloud Storage object access; bucket visibility controls Kubernetes custom metric namespace resolution. Object storage IAM is independent from the custom metrics API and Kubernetes namespace selection used by the HPA.
Question 594
A team sends high-cardinality user IDs as custom metric labels and exceeds monitoring limits with runaway time-series count. Which design should change?
- Match the custom-metric source labels, namespace, workload selectors and adapter query to the intended production Pods.
- Rename the monitored VM and assume metric-series selection is automatically corrected.
- Inspect a single average value and ignore distribution, errors and affected cohorts.
- Remove user identifiers from metric labels and aggregate by bounded service or workload dimensions.
Explanation
Correct Answer
D:High-cardinality labels multiply time series and cost; personal identifiers belong in protected traces/logs when needed, not broad metric labels.
Incorrect Answers
A:Changing an unrelated Pod setting leaves the stated cloud feature unresolved. It leaves the relevant monitoring signal or alert unaddressed.
B:Renaming a resource does not alter the actual effective configuration. The specific requirement concerns the relevant monitoring signal or alert.
C:It does not measure or notify on the signal requested by the case. The necessary distinction is the relevant monitoring signal or alert. The appropriate action here is: Remove user identifiers from metric labels and aggregate by bounded service or workload dimensions.
Question 595
A fleet alert is intended to notify the database team, but its notification channel targets a retired email distribution group. What should be tested?
- Send incidents to a maintained team escalation channel or pager rotation rather than a single static contact.
- Update and validate the alert policy's notification channels and incident escalation routing using an approved test.
- Set an unrelated CPU threshold without selecting the actual signal or metric labels.
- Increase Cloud Logging retention as a replacement for the required alert or metric definition.
Explanation
Correct Answer
B:An accurate condition does not deliver operational value when notification targets no longer reach responsible responders.
Incorrect Answers
A:Static configuration alone does not necessarily meet the named runtime requirement. The specific requirement concerns the relevant monitoring signal or alert.
C:It does not measure or notify on the signal requested by the case. The necessary distinction is the relevant monitoring signal or alert. The appropriate action here is: Update and validate the alert policy's notification channels and incident escalation routing using an approved test.
D:Cloud Logging records telemetry, rather than directly enforcing the requested runtime behavior. The case instead depends on the relevant monitoring signal or alert.
Question 596
An SLO burn-rate alert uses a threshold designed for a one-hour window on a 30-day aggregation and fails to detect fast outages. What should be configured?
- Apply an approved error-budget policy that delays risky changes or requires explicit exception and mitigation.
- Send every alert to all organization members instead of configuring intended notification channels.
- Use appropriate short- and long-window error-budget burn rate conditions for the documented service-level objective.
- Increase Cloud Logging retention as a replacement for the required alert or metric definition.
Explanation
Correct Answer
C:SLO burn rates have time-window semantics; excessive smoothing can conceal rapid error-budget consumption.
Incorrect Answers
A:It does not measure or notify on the signal requested by the case. The necessary distinction is the relevant monitoring signal or alert. The appropriate action here is: Use appropriate short- and long-window error-budget burn rate conditions for the documented service-level objective.
B:It does not measure or notify on the signal requested by the case. The case instead depends on the relevant monitoring signal or alert.
D:Cloud Logging records telemetry, rather than directly enforcing the requested runtime behavior. It leaves the relevant monitoring signal or alert unaddressed.
Question 597
A Managed Service for Prometheus query returns no metric from GKE even though Prometheus instrumentation emits it locally. What collection prerequisite should be checked?
- Verify managed collection configuration, scrape target selection and the workload's compatible metric endpoint.
- Match the custom-metric source labels, namespace, workload selectors and adapter query to the intended production Pods.
- Send every alert to all organization members instead of configuring intended notification channels.
- Set an unrelated CPU threshold without selecting the actual signal or metric labels.
Explanation
Correct Answer
A:Emitting a metric in a container is not sufficient for Google Cloud Managed Service for Prometheus to scrape it.
Incorrect Answers
B:Changing an unrelated Pod setting leaves the stated cloud feature unresolved. The case instead depends on the relevant monitoring signal or alert.
C:It does not measure or notify on the signal requested by the case. It leaves the relevant monitoring signal or alert unaddressed. The appropriate action here is: Verify managed collection configuration, scrape target selection and the workload's compatible metric endpoint.
D:It does not measure or notify on the signal requested by the case. The specific requirement concerns the relevant monitoring signal or alert.
Question 598
A dashboard shows CPU charts but misses a newly added Cloud Run revision because its filter pins an obsolete revision label. What should be updated?
- Adjust the Monitoring filter to include intended live revisions or group safely by service/revision while preserving historical context.
- Monitor tail-latency percentiles such as P95 or P99 and error rates for representative request cohorts.
- Set an unrelated CPU threshold without selecting the actual signal or metric labels.
- Disable the alerting policy so the current errors no longer appear.
Explanation
Correct Answer
A:Metric filters with exact resource labels may exclude new deployments even if telemetry collection is healthy.
Incorrect Answers
B:It does not measure or notify on the signal requested by the case. It leaves the relevant monitoring signal or alert unaddressed. The appropriate action here is: Adjust the Monitoring filter to include intended live revisions or group safely by service/revision while preserving historical context.
C:It does not measure or notify on the signal requested by the case. The specific requirement concerns the relevant monitoring signal or alert. The appropriate action here is: Adjust the Monitoring filter to include intended live revisions or group safely by service/revision while preserving historical context.
D:Removing an alert conceals the failing signal instead of resolving its cause. The necessary distinction is the relevant monitoring signal or alert.
Question 599
A storage service reports low request latency but users experience multi-second response times due to an upstream API gateway. What observability approach clarifies the delay?
- Trace the end-to-end request through gateway and backend spans using correlated request identifiers.
- Propagate the supported distributed-tracing trace context headers through the gateway and verify that backend spans share the incoming trace ID.
- Inspect a single average value and ignore distribution, errors and affected cohorts.
- Increase Cloud Logging retention as a replacement for the required alert or metric definition.
Explanation
Correct Answer
A:Service-local latency alone does not measure all stages of a user transaction; distributed tracing isolates upstream delay.
Incorrect Answers
B:It does not measure or notify on the signal requested by the case. The specific requirement concerns the requested latency measurement. The appropriate action here is: Trace the end-to-end request through gateway and backend spans using correlated request identifiers.
C:It does not measure or notify on the signal requested by the case. The necessary distinction is the requested latency measurement.
D:Cloud Logging records telemetry, rather than directly enforcing the requested runtime behavior. The case instead depends on the requested latency measurement.
Question 600
An incident responder cannot correlate an error-rate spike with a recent rollout because dashboards omit release version metadata. What should be added?
- Separate resource-level failure from a zone/regional or provider-wide service incident using authoritative scoped status and telemetry.
- Rename the monitored VM and assume metric-series selection is automatically corrected.
- Inspect a single average value and ignore distribution, errors and affected cohorts.
- Attach safe release or revision attributes to telemetry and mark deployments on performance dashboards.
Explanation
Correct Answer
D:Version metadata ties regressions to candidate releases instead of treating all production traffic as a single aggregate.
Incorrect Answers
A:Placement changes do not automatically implement the required resource-level feature. The necessary distinction is the relevant monitoring signal or alert.
B:Renaming a resource does not alter the actual effective configuration. The case instead depends on the relevant monitoring signal or alert. The appropriate action here is: Attach safe release or revision attributes to telemetry and mark deployments on performance dashboards.
C:It does not measure or notify on the signal requested by the case. It leaves the relevant monitoring signal or alert unaddressed. The appropriate action here is: Attach safe release or revision attributes to telemetry and mark deployments on performance dashboards.