Question 90
Infrastructure engineers maintain Azure ML stacks as auditable infrastructure as code. An operator is tasked to expose identifiers needed by later deployment stages without hard-coding environment-specific resource IDs. Which mechanism provides that capability?
- Deployment what-if
- Bicep parameter file
- Template outputs
- Idempotent deployment
Explanation
Correct Answer
C:Bicep outputs provide values from created or referenced resources. They connect infrastructure stages without embedding fragile literal IDs into every workflow.
Incorrect Answers
A:What-if compares declared desired state with existing resources and reports predicted changes; it is not a substitute for runtime tests of model predictions; the outstanding need is to expose identifiers needed by later deployment stages without hard-coding environment-specific resource IDs.
B:Parameters separate deployment-time settings from the shared resource graph; they support dev and production reuse without altering the template's underlying design; this approach leaves the team unable to expose identifiers needed by later deployment stages without hard-coding environment-specific resource IDs.
D:Declarative Azure Resource Manager deployments converge the specified resources toward the template state; this does not imply that arbitrary training job execution is idempotent; the requirement, by contrast, is to expose identifiers needed by later deployment stages without hard-coding environment-specific resource IDs.
Question 91
An automation team manages Azure ML assets and infrastructure through command-line workflows. The immediate design goal is to use the supported `az ml` command family for current Azure ML workspace, job, and asset operations. Which approach addresses that goal?
- Job show command
- Azure ML CLI extension v2
- Environment registration
- YAML job specification
- Job creation command
Explanation
Correct Answer
B:The Azure ML CLI v2 extension exposes modern resource definitions and commands. Older v1 command syntax should not be assumed interchangeable with a v2 job specification.
Incorrect Answers
A:The `az ml job show` command retrieves an existing job resource; it is appropriate for status inspection, not updating training code or promoting artifacts; the requirement, by contrast, is to use the supported `az ml` command family for current Azure ML workspace, job, and asset operations.
C:Registering an Azure ML environment creates a versioned asset; it is separate from installing a package interactively on one developer notebook; the specified outcome instead requires the team to use the supported `az ml` command family for current Azure ML workspace, job, and asset operations.
D:A YAML job file declares the execution configuration Azure ML needs to run a workload; it is not itself the training dataset or model binary; the implementation under review needs to use the supported `az ml` command family for current Azure ML workspace, job, and asset operations.
E:The `az ml job create` command requests job execution from the workspace; submission starts managed execution and does not automatically register a resulting model as an approved release; the requested task remains to use the supported `az ml` command family for current Azure ML workspace, job, and asset operations.
Question 92
An automation team manages Azure ML assets and infrastructure through command-line workflows. The team is required to store the training command, inputs, compute target, and environment in source control for reproducible submission. Which option is the best fit?
- YAML job specification
- Job creation command
- Environment registration
- Job show command
Explanation
Correct Answer
A:A YAML job file declares the execution configuration Azure ML needs to run a workload. It is not itself the training dataset or model binary.
Incorrect Answers
B:The `az ml job create` command requests job execution from the workspace; submission starts managed execution and does not automatically register a resulting model as an approved release; the requested task remains to store the training command, inputs, compute target, and environment in source control for reproducible submission.
C:Registering an Azure ML environment creates a versioned asset; it is separate from installing a package interactively on one developer notebook; this alternative does not achieve the requirement to store the training command, inputs, compute target, and environment in source control for reproducible submission.
D:The `az ml job show` command retrieves an existing job resource; it is appropriate for status inspection, not updating training code or promoting artifacts; the actual design criterion is to store the training command, inputs, compute target, and environment in source control for reproducible submission.
Question 93
An automation team manages Azure ML assets and infrastructure through command-line workflows. An operator is tasked to submit a declared training workload to Azure ML rather than merely validating its local YAML text. Which mechanism provides that capability?
- Job show command
- Environment registration
- Component registration
- Job creation command
Explanation
Correct Answer
D:The `az ml job create` command requests job execution from the workspace. Submission starts managed execution and does not automatically register a resulting model as an approved release.
Incorrect Answers
A:The `az ml job show` command retrieves an existing job resource; it is appropriate for status inspection, not updating training code or promoting artifacts; the actual design criterion is to submit a declared training workload to Azure ML rather than merely validating its local YAML text.
B:Registering an Azure ML environment creates a versioned asset; it is separate from installing a package interactively on one developer notebook; the outstanding need is to submit a declared training workload to Azure ML rather than merely validating its local YAML text.
C:Registering a component makes its definition addressable by name and version; the component itself is not an always-running inference endpoint; this approach leaves the team unable to submit a declared training workload to Azure ML rather than merely validating its local YAML text.
Question 94
An automation team manages Azure ML assets and infrastructure through command-line workflows. The immediate design goal is to inspect status and metadata of an existing Azure ML job without resubmitting it. Which approach addresses that goal?
- Job show command
- Component registration
- Azure ML CLI extension v2
- YAML job specification
Explanation
Correct Answer
A:The `az ml job show` command retrieves an existing job resource. It is appropriate for status inspection, not updating training code or promoting artifacts.
Incorrect Answers
B:Registering a component makes its definition addressable by name and version; the component itself is not an always-running inference endpoint; the requirement, by contrast, is to inspect status and metadata of an existing Azure ML job without resubmitting it.
C:The Azure ML CLI v2 extension exposes modern resource definitions and commands; older v1 command syntax should not be assumed interchangeable with a v2 job specification; the team would still need to inspect status and metadata of an existing Azure ML job without resubmitting it.
D:A YAML job file declares the execution configuration Azure ML needs to run a workload; it is not itself the training dataset or model binary; the specified outcome instead requires the team to inspect status and metadata of an existing Azure ML job without resubmitting it.
Question 95
An automation team manages Azure ML assets and infrastructure through command-line workflows. The team is required to publish a reusable runtime definition so other jobs can reference a stable environment version. Which option is the best fit?
- Job creation command
- Azure ML CLI extension v2
- Job show command
- Environment registration
Explanation
Correct Answer
D:Registering an Azure ML environment creates a versioned asset. It is separate from installing a package interactively on one developer notebook.
Incorrect Answers
A:The `az ml job create` command requests job execution from the workspace; submission starts managed execution and does not automatically register a resulting model as an approved release; the specified outcome instead requires the team to publish a reusable runtime definition so other jobs can reference a stable environment version.
B:The Azure ML CLI v2 extension exposes modern resource definitions and commands; older v1 command syntax should not be assumed interchangeable with a v2 job specification; the implementation under review needs to publish a reusable runtime definition so other jobs can reference a stable environment version.
C:The `az ml job show` command retrieves an existing job resource; it is appropriate for status inspection, not updating training code or promoting artifacts; the requested task remains to publish a reusable runtime definition so other jobs can reference a stable environment version.
Question 96
An automation team manages Azure ML assets and infrastructure through command-line workflows. An operator is tasked to publish a reusable pipeline step with an explicit interface instead of copying script configuration between projects. Which mechanism provides that capability?
- Environment registration
- Component registration
- Job show command
- Azure ML CLI extension v2
Explanation
Correct Answer
B:Registering a component makes its definition addressable by name and version. The component itself is not an always-running inference endpoint.
Incorrect Answers
A:Registering an Azure ML environment creates a versioned asset; it is separate from installing a package interactively on one developer notebook; this alternative does not achieve the requirement to publish a reusable pipeline step with an explicit interface instead of copying script configuration between projects.
C:The `az ml job show` command retrieves an existing job resource; it is appropriate for status inspection, not updating training code or promoting artifacts; the outstanding need is to publish a reusable pipeline step with an explicit interface instead of copying script configuration between projects.
D:The Azure ML CLI v2 extension exposes modern resource definitions and commands; older v1 command syntax should not be assumed interchangeable with a v2 job specification; this approach leaves the team unable to publish a reusable pipeline step with an explicit interface instead of copying script configuration between projects.
Question 97
A security architect designs a workspace that cannot be reached through unrestricted public paths. The immediate design goal is to give authorized workspace administrators access to Azure Machine Learning management APIs over a trusted virtual network rather than over the public workspace ingress path. Which approach addresses that goal?
- Managed virtual network egress
- Private endpoint connection approval
- Hybrid private DNS forwarding
- Workspace management private endpoint
Explanation
Correct Answer
D:A workspace private endpoint provides a private path to Azure ML workspace services. It does not automatically make independently deployed scoring endpoints private or grant managed training compute access to storage.
Incorrect Answers
A:Managed network configuration governs outbound connectivity for supported compute; it is a different traffic direction from a user's inbound access to a private workspace; this approach leaves the team unable to give authorized workspace administrators access to Azure Machine Learning management APIs over a trusted virtual network rather than over the public workspace ingress path.
B:An Azure Private Link connection in Pending state cannot carry authorized service traffic until the target owner approves it; approval does not create the endpoint, fix client DNS, or grant application-level workspace permissions; the needed capability is to give authorized workspace administrators access to Azure Machine Learning management APIs over a trusted virtual network rather than over the public workspace ingress path.
C:Conditional forwarding routes the relevant private-link DNS queries to a resolver that can answer from Azure private DNS; a virtual-network zone link alone does not configure the on-premises DNS server to forward those requests; the requirement, by contrast, is to give authorized workspace administrators access to Azure Machine Learning management APIs over a trusted virtual network rather than over the public workspace ingress path.
Question 98
A security architect designs a workspace that cannot be reached through unrestricted public paths. The team is required to authorize a pending Azure Machine Learning workspace private endpoint request submitted from another controlled network boundary. Which option is the best fit?
- Private endpoint connection approval
- Firewall allowlist
- Hybrid private DNS forwarding
- Workspace management private endpoint
- Managed virtual network egress
Explanation
Correct Answer
A:An Azure Private Link connection in Pending state cannot carry authorized service traffic until the target owner approves it. Approval does not create the endpoint, fix client DNS, or grant application-level workspace permissions.
Incorrect Answers
B:Firewall rules restrict network-level reachability to a resource; they do not replace application authentication or grant blob read permission; the specified outcome instead requires the team to authorize a pending Azure Machine Learning workspace private endpoint request submitted from another controlled network boundary.
C:Conditional forwarding routes the relevant private-link DNS queries to a resolver that can answer from Azure private DNS; a virtual-network zone link alone does not configure the on-premises DNS server to forward those requests; the implementation under review needs to authorize a pending Azure Machine Learning workspace private endpoint request submitted from another controlled network boundary.
D:A workspace private endpoint provides a private path to Azure ML workspace services; it does not automatically make independently deployed scoring endpoints private or grant managed training compute access to storage; the requested task remains to authorize a pending Azure Machine Learning workspace private endpoint request submitted from another controlled network boundary.
E:Managed network configuration governs outbound connectivity for supported compute; it is a different traffic direction from a user's inbound access to a private workspace; this alternative does not achieve the requirement to authorize a pending Azure Machine Learning workspace private endpoint request submitted from another controlled network boundary.
Question 99
A security architect designs a workspace that cannot be reached through unrestricted public paths. An operator is tasked to control how managed Azure ML compute reaches external resources from inside the service-managed network. Which mechanism provides that capability?
- Hybrid private DNS forwarding
- Outbound private endpoint rule
- Managed virtual network egress
- Firewall allowlist
Explanation
Correct Answer
C:Managed network configuration governs outbound connectivity for supported compute. It is a different traffic direction from a user's inbound access to a private workspace.
Incorrect Answers
A:Conditional forwarding routes the relevant private-link DNS queries to a resolver that can answer from Azure private DNS; a virtual-network zone link alone does not configure the on-premises DNS server to forward those requests; the requested task remains to control how managed Azure ML compute reaches external resources from inside the service-managed network.
B:A managed network private endpoint outbound rule provides private connectivity toward a specific supported target; the target must still authorize the calling identity; this alternative does not achieve the requirement to control how managed Azure ML compute reaches external resources from inside the service-managed network.
D:Firewall rules restrict network-level reachability to a resource; they do not replace application authentication or grant blob read permission; the outstanding need is to control how managed Azure ML compute reaches external resources from inside the service-managed network.
Question 100
A security architect designs a workspace that cannot be reached through unrestricted public paths. The immediate design goal is to allow managed compute to reach an approved private dependency without broad internet egress. Which approach addresses that goal?
- Private endpoint connection approval
- Workspace management private endpoint
- Hybrid private DNS forwarding
- Outbound private endpoint rule
Explanation
Correct Answer
D:A managed network private endpoint outbound rule provides private connectivity toward a specific supported target. The target must still authorize the calling identity.
Incorrect Answers
A:An Azure Private Link connection in Pending state cannot carry authorized service traffic until the target owner approves it; approval does not create the endpoint, fix client DNS, or grant application-level workspace permissions; the outstanding need is to allow managed compute to reach an approved private dependency without broad internet egress.
B:A workspace private endpoint provides a private path to Azure ML workspace services; it does not automatically make independently deployed scoring endpoints private or grant managed training compute access to storage; this approach leaves the team unable to allow managed compute to reach an approved private dependency without broad internet egress.
C:Conditional forwarding routes the relevant private-link DNS queries to a resolver that can answer from Azure private DNS; a virtual-network zone link alone does not configure the on-premises DNS server to forward those requests; the needed capability is to allow managed compute to reach an approved private dependency without broad internet egress.