View Full Cisco CCNP Automation 350-901 Exam Dumps and Practice Test Dumps.

Q281. An Ansible playbook must configure routers in batches of five so that a configuration error cannot affect all 50 routers simultaneously. Which play-level keyword should be used?

1. run_once
   2. delegate_to
   3. register
   4. serial

Correct Answer: 4. serial

Explanation: The Ansible serial keyword controls how many hosts from a play are processed at the same time. Setting serial: 5 causes Ansible to complete the relevant portion of the play for a batch of five devices before moving to the next batch. This reduces the blast radius of an automation failure and allows validation or failure-handling logic to stop later batches if necessary. run_once executes a task only once, delegate_to changes where a task executes, and register stores task results. Batch deployment is useful when rolling out potentially disruptive network changes across large inventories.

Q282. A Terraform team needs to ensure that all engineers and CI runners use provider versions compatible with the tested automation code. Which configuration is most appropriate?

1. Define provider version constraints in the Terraform configuration and retain the dependency lock file
   2. Always download the newest provider regardless of compatibility
   3. Delete the provider lock file before every pipeline
   4. Allow each engineer to choose unrelated provider versions manually

Correct Answer: 1. Define provider version constraints in the Terraform configuration and retain the dependency lock file

Explanation: Terraform provider version constraints define which versions are acceptable for the configuration, while the dependency lock file records the selected provider versions and checksums. Together, these features improve reproducibility across engineers and CI/CD environments. Automatically using the newest provider can introduce breaking behavior without testing. Removing the lock file unnecessarily can allow version selection to change between runs. Teams should update providers deliberately, review release notes, test the resulting execution plan, and commit appropriate lock-file changes through normal version-control processes. Reproducible dependencies are essential for reliable Infrastructure as Code workflows.

Q283. A RESTCONF automation client receives HTTP 412 Precondition Failed after sending an If-Match header with an earlier ETag. What is the most likely cause?

1. The API does not support HTTPS
   2. The user supplied an invalid Git branch
   3. The resource changed after the client originally retrieved it
   4. The controller is necessarily rate limiting the client

Correct Answer: 3. The resource changed after the client originally retrieved it

Explanation: An If-Match request can be used for optimistic concurrency control. The client retrieves a resource and stores its ETag, then sends that ETag when attempting an update. If another process modifies the resource first, the server’s current ETag no longer matches the client’s value and the conditional update can be rejected with HTTP 412. The client should retrieve the current resource, reevaluate its intended change, and avoid blindly overwriting the newer configuration. This approach helps prevent automation jobs from accidentally replacing updates made by another administrator or workflow.

Q284. A Python script needs to execute cleanup code whether a network API operation succeeds or raises an exception. Which language construct best fits this requirement?

1. elif
   2. finally
   3. yield only
   4. lambda only

Correct Answer: 2. finally

Explanation: A finally block executes after the associated try statement whether the protected operation succeeds or raises an exception. It is therefore appropriate for cleanup actions such as closing connections, releasing locks, deleting temporary files, or resetting transient state. Exception-specific handling can be placed in except blocks, while finally handles tasks that must occur regardless of the result. Reliable cleanup is important in network automation because a failed operation should not leave stale locks, open sessions, or temporary resources that interfere with future workflows. Context managers can provide similar cleanup behavior for resources designed to support them.

Q285. An automation service obtains a short-lived access token from an API and also receives a refresh token. What is the primary purpose of the refresh token?

1. Obtain a new access token without requiring the complete user authorization flow again
   2. Permanently disable token expiration
   3. Replace TLS encryption
   4. Grant unlimited administrator privileges automatically

Correct Answer: 1. Obtain a new access token without requiring the complete user authorization flow again

Explanation: Refresh tokens allow an authorized client to request new access tokens after the current access token expires without forcing the user through the entire authorization process each time. They are generally longer-lived and should therefore receive stronger protection than ordinary short-lived access tokens. The client must follow the authorization server’s documented refresh workflow and handle revocation or expiration correctly. Refresh tokens do not remove the need for TLS, and they do not automatically grant broader privileges than the approved authorization. Secure persistent authentication requires careful management of token scope, lifetime, storage, and rotation.

Q286. A Git repository contains a generated directory called build/ that should never be committed. Which Git mechanism should the project use?

1. git revert
   2. git bisect
   3. git cherry-pick
   4. Add the directory pattern to .gitignore

Correct Answer: 4. Add the directory pattern to .gitignore

Explanation: A .gitignore file defines patterns for files and directories that Git should normally leave untracked. Generated build output, temporary files, local virtual environments, and editor-specific files are common candidates. Ignoring generated content helps keep the repository focused on source files and reduces unnecessary merge conflicts. If a file is already tracked, adding it to .gitignore alone does not automatically remove it from version control; the team must update the index appropriately. git revert, bisect, and cherry-pick perform history-related operations and do not define persistent ignore rules.

Q287. A GitLab pipeline should run a lightweight syntax test first and avoid consuming expensive CML resources when that test fails. What pipeline design best supports this?

1. Launch CML before every other pipeline job
   2. Place syntax validation in an earlier required stage before the CML test stage
   3. Ignore syntax-test failures
   4. Run all jobs manually and independently

Correct Answer: 2. Place syntax validation in an earlier required stage before the CML test stage

Explanation: CI/CD pipelines should generally perform fast, inexpensive checks before resource-intensive integration tests. A syntax or linting stage can reject clearly invalid automation before the pipeline launches a CML topology and consumes compute resources. If the early stage fails, later required stages do not proceed. This improves feedback speed and conserves shared infrastructure. The pipeline can then progress through increasingly realistic validation, such as unit tests, CML simulation, deployment, and post-validation. Thoughtful stage ordering helps network automation teams detect simple problems early while reserving expensive test environments for changes that pass foundational checks.

Q288. A CI pipeline needs the same CML test topology to behave consistently across repeated runs. Which practice is most important?

1. Modify the topology manually before every run
   2. Preserve whatever state the previous test left behind
   3. Create the topology and baseline state from version-controlled definitions
   4. Disable cleanup and initialization

Correct Answer: 3. Create the topology and baseline state from version-controlled definitions

Explanation: Repeatable tests depend on a known initial state. If the CML topology, node definitions, links, and baseline configurations are version-controlled and recreated predictably, identical automation can be tested under comparable conditions. Reusing a lab containing unknown changes introduces state drift and makes failures harder to reproduce. A mature pipeline should create or reset the environment, run tests, preserve required results, and clean up afterward. CML helps create realistic automation test environments, but the value of simulation depends heavily on controlling the lab’s initial conditions and software assumptions.

Q289. A telemetry architecture needs network devices to establish their own streaming connections toward a collector configured on the management network. What general model does this describe?

1. Offline telemetry
   2. Static CLI polling
   3. Git-triggered collection
   4. Device-initiated or dial-out telemetry

Correct Answer: 4. Device-initiated or dial-out telemetry

Explanation: In a device-initiated, or dial-out, telemetry architecture, the network device is configured with the destination collector and establishes the telemetry connection toward it. This affects firewall rules, reachability requirements, authentication, certificate design, and failure handling. Another model allows collectors to initiate subscriptions toward devices. Neither approach is universally best; the architecture depends on platform capabilities and operational requirements. Telemetry design should also define sensor paths, subscription behavior, encoding, update frequency, collector redundancy, and storage. Understanding who initiates the connection is important when designing secure and scalable telemetry systems.

Q290. A monitoring platform receives an extremely short CPU spike lasting two seconds. The telemetry stream samples CPU only once every five minutes. What is the primary concern?

1. Git cannot store the measurement
   2. The sampling interval may miss short-lived events
   3. YANG cannot represent CPU data
   4. TLS prevents frequent sampling

Correct Answer: 2. The sampling interval may miss short-lived events

Explanation: Telemetry sampling frequency determines the time resolution of collected data. A five-minute interval may be sufficient for long-term capacity trends but can completely miss a transient two-second CPU spike. Increasing sampling frequency improves visibility into short-lived events but also increases device processing, bandwidth, collector workload, and storage consumption. Telemetry architecture must therefore balance resolution with scalability and operational value. In some cases, event-driven or on-change mechanisms can complement periodic sampling. Engineers should choose collection behavior according to the phenomena they need to observe rather than selecting one interval for every metric.

Q291. An automation application must send an alert when a job fails, but failure of the notification system must not cause the original configuration transaction to be reported as successful. What should the application do?

1. Track the configuration result and notification result separately and report them accurately
   2. Treat notification success as proof the configuration succeeded
   3. Hide the configuration result if notification fails
   4. Retry the configuration automatically whenever the alert system is unavailable

Correct Answer: 1. Track the configuration result and notification result separately and report them accurately

Explanation: Configuration execution and notification delivery are separate operations with different meanings. A failed notification does not change whether a router configuration succeeded, and a successful alert does not prove the network change worked. Automation should record each result independently and expose enough context for operators to understand both conditions. If notification is important, it can use bounded retries or an alternate channel without altering the factual deployment status. Separating side-effect failures from the core transaction makes automation state clearer and prevents operational dashboards from presenting misleading conclusions.

Q292. A pyATS test compares interface counters before and after a load test. Which practice helps prevent legitimate counter increases from being reported as configuration failures?

1. Require all counters to remain exactly identical
   2. Remove counter collection entirely
   3. Define expected ranges or comparison logic appropriate to the metric
   4. Mark every changing value as failed

Correct Answer: 3. Define expected ranges or comparison logic appropriate to the metric

Explanation: Not every operational metric should be tested using strict equality. Traffic counters are expected to increase during normal activity, so validation should reflect the meaning of the metric. A test might verify that error counters stay below a threshold, that packet counts increase, or that utilization remains within a permitted range. Well-designed automation distinguishes expected state changes from genuine failures. pyATS can provide structured values, but engineers must still define meaningful assertions. Tests that use inappropriate comparisons generate false failures and reduce confidence in the automated validation system.

Q293. A network automation API runs in a container. The team wants the service to execute as a non-root user even if the container runtime allows root. What security principle does this implement?

1. Least privilege
   2. High availability
   3. Event correlation
   4. Pagination

Correct Answer: 2. Least privilege

Explanation: Running an application with only the permissions it requires follows the principle of least privilege. If the automation service is compromised, a non-root runtime identity can reduce the attacker’s ability to modify the container or interact with host resources. Container security should also consider filesystem permissions, capabilities, image provenance, secrets, networking, and runtime isolation. Root inside a container is not automatically equivalent to unrestricted host root access, but unnecessarily elevated permissions still increase risk. Secure automation design should minimize privileges at the application, container, API, and infrastructure layers.

Q294. An internal automation service uses HTTPS, but a client must connect through the service’s IP address instead of its DNS name. The certificate contains only the DNS name in its SAN. What is likely to happen with normal certificate validation?

1. Validation always succeeds because the certificate is signed
   2. The hostname or identity check fails because the IP is not an approved certificate identity
   3. TLS automatically adds the IP to the certificate
   4. The client receives the server private key

Correct Answer: 4. The hostname or identity check fails because the IP is not an approved certificate identity

Explanation: TLS identity validation compares the address used by the client with identities listed in the certificate’s Subject Alternative Name extension. A certificate containing only a DNS name normally does not validate when the client connects using an unrelated IP address, even if the certificate chain itself is trusted. If clients legitimately connect using an IP address, that identity must be considered during certificate issuance where supported and appropriate. Disabling hostname verification is generally an unsafe workaround because it weakens protection against impersonation and man-in-the-middle attacks.

Q295. A local LLM is too large to run efficiently on available hardware. The team wants to reduce memory requirements while retaining the same general model. Which technique should it evaluate?

1. Quantization
   2. Increasing every parameter value
   3. Removing all network data validation
   4. Increasing the model size further

Correct Answer: 1. Quantization

Explanation: Quantization represents model weights using lower numerical precision, which can substantially reduce memory consumption and sometimes improve inference speed. This can make a local LLM practical on less powerful hardware. The tradeoff is that aggressive quantization may reduce model quality, so the team should benchmark technical accuracy and latency on representative network automation tasks. Quantization does not make an AI system safer by itself and does not replace testing, authorization, retrieval controls, or tool validation. Local LLM deployment should balance model quality, hardware capacity, response time, and data-governance requirements.

Q296. A retrieval-augmented network assistant should answer only from approved documentation and explicitly state when supporting information cannot be found. What prompt strategy best supports this behavior?

1. Tell the model to invent missing details confidently
   2. Instruct the model to ground answers in supplied evidence and acknowledge when evidence is insufficient
   3. Remove all retrieved context
   4. Increase randomness to maximize creativity

Correct Answer: 3. Instruct the model to ground answers in supplied evidence and acknowledge when evidence is insufficient

Explanation: Clear system instructions can direct a retrieval-augmented assistant to use the supplied evidence, distinguish evidence from inference, and state when available documents do not support an answer. This helps reduce unsupported claims, although prompt instructions alone cannot guarantee correctness. The retrieval layer should also supply authoritative, current, and authorized documents, and the application can preserve citations or provenance for verification. Deterministic validation remains necessary for operational actions. For network troubleshooting, an explicit “insufficient evidence” result is safer than confidently inventing configuration details that are not present in approved sources.

Q297. An MCP server exposes a tool that normally completes in two seconds but occasionally hangs because of an upstream controller failure. Which server behavior is most appropriate?

1. Wait forever so the model never receives an error
   2. Allow unlimited parallel hanging requests
   3. Restart every router when the request stalls
   4. Enforce a bounded timeout and return a controlled tool error

Correct Answer: 4. Enforce a bounded timeout and return a controlled tool error

Explanation: MCP tools depend on external systems that can become slow or unavailable. The server should set explicit timeouts so one controller problem cannot block an AI workflow indefinitely or consume server resources permanently. The error returned to the agent should accurately state that current data could not be obtained rather than encouraging the model to fabricate a result. Depending on the operation, bounded retry with backoff may be appropriate. Timeouts, concurrency controls, input validation, authorization, and logging are important safeguards when probabilistic AI agents interact with deterministic network tools.

Q298. An AI agent suggests a router change based on one telemetry source, but a second independent monitoring system reports contradictory state. What should the workflow do before changing production?

1. Reconcile the conflicting evidence or escalate for validation rather than assuming one source is correct
   2. Always trust the AI’s preferred source
   3. Ignore both monitoring systems
   4. Apply the change first and investigate afterward

Correct Answer: 2. Reconcile the conflicting evidence or escalate for validation rather than assuming one source is correct

Explanation: Conflicting operational evidence creates uncertainty that should be resolved before an impactful change is executed. The workflow can query the network directly, evaluate source freshness and authority, run deterministic validation, or escalate the discrepancy to an operator. An AI model should not arbitrarily decide that one data source is correct merely because it appears first or is easier to interpret. Production automation should explicitly handle uncertainty because acting on incorrect telemetry can turn an observability problem into an outage. Evidence reconciliation is particularly important for AI-assisted diagnosis and remediation.

Q299. An AI coding assistant generates a network automation dependency that refers to a package name nearly identical to a legitimate internal package. What risk should the development team consider?

1. DHCP exhaustion
   2. Certificate expiration
   3. Software supply-chain or dependency-confusion risk
   4. Model-driven telemetry sampling

Correct Answer: 4. Software supply-chain or dependency-confusion risk

Explanation: AI-generated code can reference incorrect, nonexistent, or similarly named dependencies. If developers install an unverified package from a public repository, they can introduce malicious code into the automation environment. Teams should verify dependency names against authoritative project documentation, use approved repositories, pin or constrain versions appropriately, and scan dependencies before use. Internal package naming should also be designed to reduce dependency-confusion opportunities. Generative AI can accelerate development, but its suggested libraries must be treated as untrusted until they are verified through normal software supply-chain controls.

Q300. A network AI agent automatically performs read-only diagnostics but creates a proposed change ticket whenever remediation would modify production. What operating model does this best represent?

1. Complete autonomous control
   2. Anonymous network administration
   3. Tiered autonomy based on action impact
   4. Removal of human oversight for every action

Correct Answer: 3. Tiered autonomy based on action impact

Explanation: Tiered autonomy gives an AI agent different levels of independence according to operational risk. Read-only diagnostics can often be performed automatically because they have limited direct impact, while configuration changes can require a proposal, ticket, approval, or separate deployment workflow. This architecture provides useful automation without granting the model unrestricted control over infrastructure. Deterministic authorization should enforce the distinction technically rather than depending only on prompt instructions. Logging, source-of-truth checks, prevalidation, rollback planning, and post-validation can provide additional safeguards when an approved change eventually reaches production.