Professional Cloud DevOps Engineer Practice Test

Answer choices

1. A. Send the same alert twice to the same Slack channel with different display names, for the same production alert, responder ownership, escalation policy, and active on-call coverage window.
2. B. Disable incident notifications and rely on engineers watching the Monitoring dashboard continuously.
3. C. Configure a redundant notification channel using a different delivery mechanism, such as email or Pub/Sub, in addition to Slack.
4. D. Create two Slack channels because duplicate destinations always use independent delivery infrastructure.

Wrong-answer review

• A. Send the same alert twice to the same Slack channel with different display names, for the same production alert, responder ownership, escalation policy, and active on-call coverage window.: Duplicate messages through the same mechanism do not create independent delivery redundancy.
• B. Disable incident notifications and rely on engineers watching the Monitoring dashboard continuously.: Manual dashboard watching is not a reliable substitute for incident notification.
• D. Create two Slack channels because duplicate destinations always use independent delivery infrastructure.: Two Slack destinations can still share the same underlying notification service.

Answer choices

1. A. Change the log severity to ERROR because sinks route only ERROR and higher by default.
2. B. Increase the source log bucket retention because sink delivery stops when source retention is shorter than the destination.
3. C. Check the sink writer identity and confirm it has the destination-specific IAM permission required to write there.
4. D. Create more Ops Agent instances because sinks require an agent on the destination project.

Answer choices

1. A. Keep all job state only on each Spot VM's local memory because reclaimed instances preserve memory for replacement VMs.
2. B. Assign each job permanently to one Spot VM and disable retry so duplicate work never occurs.
3. C. Store the only copy of each input on the Spot VM boot disk and delete upstream data after task assignment.
4. D. Use checkpointed or idempotent tasks backed by durable queues or state so reclaimed Spot workers can be replaced and work retried safely.

Answer choices

1. A. Rely on autoscaling because managed services ignore project quotas during user-facing traffic spikes.
2. B. Wait for the quota to be hit in production and then open a support case while requests are failing.
3. C. Set the autoscaler maximum above the quota because configuration values automatically reserve additional quota.
4. D. Review and obtain needed quota increases or redesign capacity before relying on autoscaling for the event.

Answer choices

1. A. Provision each preview environment from versioned infrastructure and application configuration, then automate teardown from the pull-request lifecycle.
2. B. Keep one shared staging namespace and let every pull request overwrite the same resources for faster feedback.
3. C. Clone the current production project manually for each pull request and delete it when an engineer remembers to clean up, across the same fleet-managed cluster environments while retaining the environment boundaries and regional ownership model described in the scenario.
4. D. Give each developer a permanent production-like project and stop updating it after initial provisioning.

Answer choices

1. A. Allow uploads from administrators and assume IAM administrator status proves the artifact passed CI tests.
2. B. Restrict production repository write access to the trusted build identity and enforce deploy-time provenance or attestation policy.
3. C. Disable repository audit logging so manual uploads cannot confuse the automated deployment history.
4. D. Keep direct uploads but require engineers to add the same image tag format used by Cloud Build, at the production deployment gate for images produced by the organization's approved build, provenance, and attestation path.

Answer choices

1. A. Use Network Intelligence Center Performance Dashboard to compare project network observations with Google Cloud network performance.
2. B. Use Cloud Build history because worker duration is a direct measurement of production VPC latency, for the same workload, region, production traffic profile, and representative measurement interval used to establish the performance baseline.
3. C. Use Error Reporting because every network latency spike generates an application exception group.
4. D. Use Artifact Registry because regional image replication metrics reveal live packet loss between workloads.

Answer choices

1. A. Measure only the single slowest request each month because one outlier defines overall user experience.
2. B. Keep average latency because averages always capture tail behavior more accurately than request-based thresholds.
3. C. Use total CPU time as the latency SLI because CPU is the main source of response delay.
4. D. Define an SLI based on the proportion of requests meeting a user-relevant latency threshold rather than relying only on the mean.

Answer choices

1. A. Use a dedicated service account for the automation and grant only the required roles on the production project or narrower resources.
2. B. Create a service account key, store it in the source repository, and rotate the key during each release, within the same least-privilege identity boundary and resource scope required by the workload or pipeline.
3. C. Run the automation with a developer's user credentials because user identities are easier to audit.
4. D. Create one organization-wide service account with Owner so every automation job can reuse the same identity.

Answer choices

1. A. Delete completed rollout records after 30 days and reconstruct deployment history from source-control timestamps.
2. B. Preserve immutable artifact references and correlated build, release, rollout, and audit metadata instead of relying on mutable labels.
3. C. Use the latest image tag for every deployment because the tag always points to the most recently approved artifact, for images produced by the same approved builder and evaluated at the production supply-chain policy gate.
4. D. Record only the human-friendly release name because names are sufficient to reproduce any historical artifact.