A. Incorrect.
Secret Manager environment variables are resolved for container instances and are not the best fit when the application must see secret rotations quickly without replacing instances.

B. Correct.
Mounting the Secret Manager secret as a volume lets Cloud Run expose newer secret versions to the running service without baking credentials into code or images.

C. Incorrect.
This couples credential rotation to image rebuilds and redeployments, increasing downtime and operational risk.

D. Incorrect.
Secrets must not be stored directly in source code, and rebuilds for every rotation do not meet secure DevOps practices.

A. Correct.
Personalized Service Health annotations overlay relevant Google Cloud service events on Monitoring charts, making it easier to correlate symptoms with provider-side incidents.

B. Incorrect.
Alerting on application errors detects symptoms, but it does not identify whether a Google Cloud service incident is affecting your project.

C. Incorrect.
Log-based metrics can be useful, but they are indirect and require you to infer whether the source is your application or a cloud service issue.

D. Incorrect.
Raw logs help investigation but do not provide the curated Google Cloud service-health context requested.

A. Correct.
Workload Identity lets GKE workloads use short-lived credentials and impersonate narrowly scoped Google service accounts without key files.

B. Incorrect.
Node service accounts are too coarse-grained for per-workload least privilege and become hard to manage in multi-tenant clusters.

C. Incorrect.
Service account keys are long-lived credentials and increase leakage and rotation risk.

D. Incorrect.
A shared broad service account violates least privilege and makes workload-level accountability difficult.

A. Incorrect.
Profiler helps with CPU and memory hot paths inside services, but it does not directly show distributed request latency between services.

B. Correct.
Cloud Trace is designed for distributed tracing, so it can show request spans and latency across downstream HTTP calls.

C. Incorrect.
Debugging is not the right primary observability tool for dependency latency analysis.

D. Incorrect.
Frontend-only duration logs do not show which downstream service or span is responsible for slow requests.

A. Incorrect.
Periodic tests delay feedback and can miss pull-request-specific validation before review.

B. Incorrect.
Testing after merge is too late because untested code can already enter the protected branch.

C. Incorrect.
Manual reviewer execution is inconsistent and does not provide a reliable automated gate.

D. Correct.
Cloud Build GitHub pull request triggers provide automated pre-merge testing, and build steps can run tests from the correct module directory.

A. Incorrect.
Pull request triggers validate proposed changes, but the requirement is that production image builds run only after code is pushed to main.

B. Incorrect.
Included-files filters do not enforce approval for all pushes to the main branch.

C. Correct.
A push-to-branch trigger scoped to main runs the production image build only when code reaches the approved branch.

D. Correct.
Branch protection can require reviews or approvals before changes are merged or pushed to main.

E. Incorrect.
Manual builds reduce automation and are not necessary when branch protection and scoped triggers can enforce the workflow.

A. Correct.
Node pool maximums must allow enough capacity for growth and for redistributing Pods if a zone fails.

B. Correct.
HPA helps workloads scale based on demand and keeps capacity aligned with usage patterns.

C. Correct.
Load testing validates assumptions about CPU, memory, startup time, and behavior during zone loss.

D. Incorrect.
Fixed node counts make growth and failure recovery harder and can cause under-provisioning.

E. Incorrect.
Concentrating Pods in one zone reduces availability and conflicts with the zone-failure requirement.

A. Correct.
Cloud Monitoring exposes Kubernetes ephemeral storage metrics that can be queried without exec or SSH access.

B. Incorrect.
This requires exec access to Pods, which the scenario explicitly says you do not have.

C. Incorrect.
This requires node access and is not necessary when monitoring metrics are available.

D. Incorrect.
Deleting Pods is a disruptive action and does not identify the root cause safely.

A. Incorrect.
Logging Editor is broader than necessary because the instances only need to write log entries.

B. Incorrect.
Admin and editor permissions are overbroad and violate least privilege.

C. Incorrect.
Monitoring Editor allows managing monitoring resources; writing metrics only needs metricWriter.

D. Correct.
logging.logWriter and monitoring.metricWriter grant the minimum write permissions needed for logs and metrics ingestion.

A. Incorrect.
Load tests are useful, but waiting for a prototype delays operational feedback.

B. Incorrect.
Operational requirements should be considered before QA and staging, not only afterward.

C. Incorrect.
Waiting for production usage increases customer risk and makes design changes more expensive.

D. Correct.
SRE input is most valuable early, when architecture, SLOs, observability, scalability, and cost controls can still be designed into the product.