Troubleshooting practice questions

You are troubleshooting a model deployment failure where the container fails to start due to missing environment variables. Where do you find the logs to identify the cause?

Trap 1: The Key Vault access logs.

Key Vault logs track secret access, not container startup.

Trap 2: The pipeline run history.

Pipeline history stores training logs, not deployment logs.

Trap 3: The workspace diagnostic logs in Azure Monitor.

These show general resource health, not app-specific logs.

A
The 'get-logs' command for the online deployment.
This retrieves the specific container startup error logs.
B
The Key Vault access logs.
Why wrong: Key Vault logs track secret access, not container startup.
C
The pipeline run history.
Why wrong: Pipeline history stores training logs, not deployment logs.
D
The workspace diagnostic logs in Azure Monitor.
Why wrong: These show general resource health, not app-specific logs.

Question 2hardmulti select

You are troubleshooting a failed agent deployment. Which TWO areas should you inspect first?

Trap 1: The Azure status page.

Rarely useful for specific project-level issues.

Trap 2: The user profile settings.

User settings do not impact resource deployment.

Trap 3: The billing usage report.

Billing is not relevant to deployment success.

A
The Azure status page.
Why wrong: Rarely useful for specific project-level issues.
B
The endpoint availability status.
Indicates if the infrastructure is actually accepting requests.
C
The 'Deployment Logs' in the Azure AI Foundry portal.
Logs show the underlying error during provisioning.
D
The user profile settings.
Why wrong: User settings do not impact resource deployment.
E
The billing usage report.
Why wrong: Billing is not relevant to deployment success.

Question 3mediummulti select

A developer is troubleshooting an agent that is failing to connect to an external tool. Which THREE of the following infrastructure settings should be verified?

Trap 1: The Azure Resource Group location.

The region of the resource group is unlikely to cause a specific connectivity failure to an external tool.

Trap 2: The model version of the LLM.

The LLM model version affects reasoning capability, not the network connectivity to external tools.

A
The Azure AI Foundry 'Connections' configuration for the specific tool.
The Connections object manages the credentials/API keys for external services.
B
The Environment Variables passed to the Prompt Flow runtime.
Environment variables often contain the endpoint URLs or credentials required for external tool integration.
C
The Azure Resource Group location.
Why wrong: The region of the resource group is unlikely to cause a specific connectivity failure to an external tool.
D
The model version of the LLM.
Why wrong: The LLM model version affects reasoning capability, not the network connectivity to external tools.
E
The Network Security Group (NSG) rules allowing outbound traffic from the runtime.
NSG rules can block the agent from reaching external APIs.

Question 4hardmulti select

You are troubleshooting a deployment. Which THREE logs are most helpful to check?

Trap 1: User browser logs.

Irrelevant to the model.

Trap 2: Azure billing logs.

Irrelevant to the model.

A
Deployment status logs.
Shows infrastructure issues.
B
Application output logs (stdout).
Shows code errors.
C
User browser logs.
Why wrong: Irrelevant to the model.
D
Container build logs.
Shows image creation issues.
E
Azure billing logs.
Why wrong: Irrelevant to the model.

Question 5hardmultiple choice

You are troubleshooting high latency in a RAG-based application. The vector search is fast, but the generation phase is slow. Which component should be scaled?

Trap 1: Optimize the vector database index.

The issue is in generation, not the vector retrieval phase.

Trap 2: Reduce the 'temperature' setting.

Temperature impacts response quality, not inference latency.

Trap 3: Increase the 'Embedding Model' throughput.

The bottleneck is in generation, not embedding.

A
Optimize the vector database index.
Why wrong: The issue is in generation, not the vector retrieval phase.
B
Increase the 'Instances' count for the LLM deployment.
Scaling the LLM inference instance count increases generation throughput.
C
Reduce the 'temperature' setting.
Why wrong: Temperature impacts response quality, not inference latency.
D
Increase the 'Embedding Model' throughput.
Why wrong: The bottleneck is in generation, not embedding.