GCP Professional Cloud Architect Practice Question

After a new microservice release, your company's checkout API running on GKE intermittently takes five seconds to respond instead of the usual 200 ms. Logs alone are too noisy to pinpoint the issue. You need to quickly locate which downstream call causes the spike and inspect local variables at that moment without restarting any pods. Which Google Cloud-based troubleshooting workflow best follows RCA best practices?

Run an on-demand CPU profile with Cloud Profiler, export the profile to BigQuery, and analyze blocking threads to infer which call causes latency.
Filter slow requests in Cloud Trace, open the correlated logs from the trace viewer, then set a non-breaking snapshot in Cloud Debugger on the suspicious downstream call to capture its state during the next spike.
Add latency guard unit tests to the CI/CD pipeline and automatically roll back the release if tests exceed five-second response times.
Raise log verbosity to debug on every container, collect 24 hours of logs into BigQuery with Dataflow, and query for the average latency of each function.

Report Issue

Answer Description

The most efficient root-cause analysis starts by identifying the slow requests in Cloud Trace, because each trace shows a detailed timeline of every span across services. From a problematic trace, the View logs link lets you jump directly to the set of log entries generated during that request, eliminating the noise of unrelated traffic. Once the slow span is identified, Cloud Debugger snapshots can be set on the running container to capture stack frames and variable values the next time the code path is executed, all without stopping or redeploying the service. This end-to-end path-Trace ➜ correlated Logs ➜ non-breaking Debugger snapshot-follows Google Cloud's recommended RCA process of reproducing, isolating, and inspecting a live failure. The other options either rely on coarse profiling data, delayed log analysis, or pre-deployment testing, none of which directly correlate individual slow requests with in-process state in production.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.