2.2 Implementing and Managing Pipelines PROFESSIONAL-CLOUD-DEVOPS-ENGINEER Practice Questions (Page 4)

Questions 16–20

1. 16expert · hard

   You are deploying a new version of a recommendation model. The model is served through a Cloud Run service. You have set up a canary deployment where 5% of traffic goes to the new model version. You need to define a success metric to determine whether to proceed with the full rollout. The model's offline AUC is 0.85. Which metric is the most appropriate to use as the primary success metric for the canary deployment?

   • AThe click-through rate (CTR) on recommendations served by the new model compared to the old model.
   • BThe model's AUC calculated on a sample of production traffic with human-labeled feedback.
   • CThe model's AUC on the offline validation dataset.
   • DThe model's prediction latency at the 99th percentile.
   Reveal Solution Discussion Select an answer first
2. 17expert · hard

   You are deploying a new version of a machine learning model that detects fraudulent transactions. The model is served through a Cloud Run service. You have set up a canary deployment where 5% of traffic goes to the new model. You need to define a success metric and an automated rollback criterion. The model's offline F1 score is 0.92. Which combination of success metric and rollback criterion is the most appropriate?

   • ASuccess metric: Model prediction latency. Rollback criterion: Roll back if the 99th percentile latency exceeds 500ms.
   • BSuccess metric: The model's prediction accuracy on the offline validation dataset. Rollback criterion: Roll back if the accuracy drops below 0.95.
   • CSuccess metric: Model F1 score on a sample of production data with human-labeled feedback. Rollback criterion: Roll back if the F1 score drops below 0.90.
   • DSuccess metric: The rate of false positives (legitimate transactions flagged as fraudulent). Rollback criterion: Roll back if the false positive rate exceeds 1%.
   Reveal Solution Discussion Select an answer first
4. 18foundation · easy

   What is a key consideration when deploying a new version of a machine learning model?

   • ADisabling data drift monitoring
   • BModel versioning to track which model is serving predictions
   • CDeleting old model versions
   • DIncreasing the number of GPUs
   Reveal Solution Discussion Select an answer first
5. 19application · medium

   You are configuring an automated rollout for a new version of your application using Cloud Deploy. You want to automatically roll back the deployment if the error rate exceeds 5% for more than 5 minutes. You have set up a Cloud Monitoring alerting policy that triggers when the error rate exceeds this threshold. What is the next step to enable automated rollback?

   • AConfigure the GKE Deployment's 'rollbackOnError' field in the manifest.
   • BUse the Cloud Deploy '--rollback-on-error' flag when creating the rollout.
   • CConfigure the Cloud Deploy rollout to watch the Cloud Monitoring alert and automatically roll back when the alert fires.
   • DCreate a Cloud Function that is triggered by the Cloud Monitoring alert and calls the Cloud Deploy API to roll back the rollout.
   Reveal Solution Discussion Select an answer first
6. 20foundation · easy

   What is the purpose of defining a threshold for automated rollback in a deployment pipeline?

   • ATo log all deployment actions
   • BTo scale up resources automatically
   • CTo automatically revert to the previous version if telemetry signals exceed the threshold
   • DTo prevent any deployment from happening
   Reveal Solution Discussion Select an answer first