100% Latest & Updated Google Professional Cloud Architect Practice Test Questions, Exam Dumps & Verified Answers!
30 Days Free Updates, Instant Download!
Question 320
A migration program deploys new data services but cannot safely switch over until the old and new workloads produce equivalent outputs. Which operational action best maintains postcutover discrepancy response after launch?
- Declare cutover finished immediately after DNS change
- Disable reconciliation alerts once new traffic reaches target
- Maintain a short reconciliation window and ownership of mismatch remediation
- Revert to source despite independent target writes without a reconciliation plan
Explanation
Correct Answer
C:A completed traffic switch can hide data errors that emerge later
Incorrect Answers
A:Ignores delayed mismatches. A completed traffic switch can hide data errors that emerge later.
B:Can conceal data defects. A completed traffic switch can hide data errors that emerge later.
D:May lose accepted data. A completed traffic switch can hide data errors that emerge later.
Question 321
A machine-learning inference application is deployed behind Cloud Run and its new model is slower for a subset of customer cohorts. Which architecture most directly meets the need for cohort-aware serving rollout?
- Use a controlled canary with model-quality and latency monitoring per cohort
- Promote the new model to all cohorts with no comparison
- Use only a global average latency report
- Choose model release solely by file size
Explanation
Correct Answer
A:Aggregate latency can hide regressions affecting important customer groups
Incorrect Answers
B:May harm a small sensitive group. Aggregate latency can hide regressions affecting important customer groups.
C:Can hide regressions in minority cohorts. Aggregate latency can hide regressions affecting important customer groups.
D:Not serving quality. Aggregate latency can hide regressions affecting important customer groups.
Question 322
A machine-learning inference application is deployed behind Cloud Run and its new model is slower for a subset of customer cohorts. Which implementation choice most directly delivers governed model version promotion?
- Overwrite model artifact files under a mutable name during rollout
- Permit unreviewed model versions to receive production traffic
- Store traffic split rules outside the release record
- Pin approved model artifact versions and record their deployment-to-traffic mapping
Explanation
Correct Answer
D:A mutable model alias makes it difficult to reproduce observed predictions
Incorrect Answers
A:Breaks reproducibility. A mutable model alias makes it difficult to reproduce observed predictions.
B:Bypasses approval. A mutable model alias makes it difficult to reproduce observed predictions.
C:Cannot trace affected cohorts. A mutable model alias makes it difficult to reproduce observed predictions.
Question 323
A machine-learning inference application is deployed behind Cloud Run and its new model is slower for a subset of customer cohorts. Which validation would best establish proof of cohort safety?
- Compare one aggregate request latency average
- Test only offline model accuracy and skip live serving behavior
- Check endpoint uptime without labeled output evaluation
- Run labeled inference and request-latency tests for the affected groups
Explanation
Correct Answer
D:A global model benchmark may mask quality and service regressions in smaller cohorts
Incorrect Answers
A:Not p95 by cohort. A global model benchmark may mask quality and service regressions in smaller cohorts.
B:Misses real latency. A global model benchmark may mask quality and service regressions in smaller cohorts.
C:Cannot establish quality. A global model benchmark may mask quality and service regressions in smaller cohorts.
Question 324
A machine-learning inference application is deployed behind Cloud Run and its new model is slower for a subset of customer cohorts. Which operational action best maintains model rollback incident action after launch?
- Increase model autoscaling during a cohort-specific prediction defect
- Delay rollback until next model training cycle
- Ignore model-version labels in incident logs
- Route affected traffic back to a known-good artifact when cohort guardrails fail
Explanation
Correct Answer
D:Immediate model rollback can limit harmful predictions while engineers investigate
Incorrect Answers
A:May not fix wrong outputs. Immediate model rollback can limit harmful predictions while engineers investigate.
B:Extends harm. Immediate model rollback can limit harmful predictions while engineers investigate.
C:Prevents precise impact analysis. Immediate model rollback can limit harmful predictions while engineers investigate.
Question 325
A public content service uses old deployment scripts and operations cannot tell which code is currently receiving traffic. Which architecture most directly meets the need for visible release management?
- Deploy source code using an unversioned script with no artifact record
- Deploy versioned immutable build artifacts with an explicit release history
- Treat a mutable tag as immutable release identity
- Allow operators to copy binaries to VM disks without change tracking
Explanation
Correct Answer
B:Immutable versions map running behavior to a specific reviewed source revision
Incorrect Answers
A:Cannot map running bytes to reviewed source. Immutable versions map running behavior to a specific reviewed source revision.
C:Tags can move. Immutable versions map running behavior to a specific reviewed source revision.
D:Introduces unreviewed drift. Immutable versions map running behavior to a specific reviewed source revision.
Question 326
A public content service uses old deployment scripts and operations cannot tell which code is currently receiving traffic. Which implementation choice most directly delivers automated deployment metadata?
- Write only the deployment timestamp but no artifact digest
- Attach only a generic issue ticket not linked to deployment
- Permit every developer to modify release history logs
- Tag each release with artifact digest, change request and rollback target
Explanation
Correct Answer
D:Traceable release metadata supports incident diagnosis after multiple deploys
Incorrect Answers
A:Does not identify changed code. Traceable release metadata supports incident diagnosis after multiple deploys.
B:Insufficient provenance. Traceable release metadata supports incident diagnosis after multiple deploys.
C:Weakens audit accountability. Traceable release metadata supports incident diagnosis after multiple deploys.
Question 327
A public content service uses old deployment scripts and operations cannot tell which code is currently receiving traffic. Which validation would best establish proof of release provenance?
- Inspect repository HEAD and assume it is running
- Verify a ticket is closed but do not query live revision
- Check service health but ignore actual deployed digest
- Query the live service revision and reconcile it to the approved build digest
Explanation
Correct Answer
D:A ticket marked closed does not identify the bytes actually serving requests
Incorrect Answers
A:Production may use different commit. A ticket marked closed does not identify the bytes actually serving requests.
B:Administrative status not binary evidence. A ticket marked closed does not identify the bytes actually serving requests.
C:Not release provenance. A ticket marked closed does not identify the bytes actually serving requests.
Question 328
A public content service uses old deployment scripts and operations cannot tell which code is currently receiving traffic. Which operational action best maintains operational release forensics after launch?
- Correlate incident onset with deployment timestamps and quickly revert a suspected regression
- Leave a known failing revision active while only documenting the incident
- Retain no previous approved image for rollback
- Treat all performance regressions as unrelated to releases despite timing correlation
Explanation
Correct Answer
A:Version visibility shortens mean time to mitigate application releases
Incorrect Answers
B:Does not mitigate. Version visibility shortens mean time to mitigate application releases.
C:Slows recovery. Version visibility shortens mean time to mitigate application releases.
D:Misses a likely cause. Version visibility shortens mean time to mitigate application releases.
Question 329
A payment application team must coordinate migration tooling and application releases across ten regional sites. Which architecture most directly meets the need for coordinated distributed deployment?
- Upgrade all sites simultaneously without dependency pilots
- Use a staged rollout plan with prerequisite checks for regions and their network/data dependencies
- Choose rollout sequence randomly by alphabetical site name
- Use network uptime as the sole migration readiness factor
Explanation
Correct Answer
B:Regional waves reduce simultaneous failure exposure and missing dependency surprises
Incorrect Answers
A:One failure can affect every region. Regional waves reduce simultaneous failure exposure and missing dependency surprises.
C:Ignores readiness differences. Regional waves reduce simultaneous failure exposure and missing dependency surprises.
D:Not application/data dependencies. Regional waves reduce simultaneous failure exposure and missing dependency surprises.
Question 330
A payment application team must coordinate migration tooling and application releases across ten regional sites. Which implementation choice most directly delivers explicit wave cutover gates?
- Promote a site before rollback is tested
- Require a tested rollback path and responsible regional owner before each promotion
- Assign one global release owner without local support contacts
- Ignore site-specific maintenance windows and legal cutover restrictions
Explanation
Correct Answer
B:A global calendar is not enough to ensure each location is actually ready
Incorrect Answers
A:Increases unrecoverable impact. A global calendar is not enough to ensure each location is actually ready.
C:Regional handover gaps remain. A global calendar is not enough to ensure each location is actually ready.
D:Business risk. A global calendar is not enough to ensure each location is actually ready.
Google Cloud Certified - Professional Cloud Architect (Professional Cloud Architect) Practice Tests