Red Hat Certified Specialist in OpenShift AI (EX267) Practice Test – 1370 Free Exam Questions with Answers

Correct answer: C. Red Hat OpenShift AI Operator

The Red Hat OpenShift AI Operator handles the installation and lifecycle management of the platform components. The OpenShift Data Science Operator is the legacy name from older versions and is no longer the correct terminology.

Correct answer: D. Using Environment Variables sourced from Secrets or ConfigMaps

Environment variables sourced from Secrets or ConfigMaps provide secure configuration injection for running workbenches. Hardcoding credentials or storing them in plain text violates basic security best practices.

Correct answer: C. Bias / Fairness

TrustyAI evaluates bias and fairness by monitoring model outcomes for demographic imbalances. Concept drift refers to changing data patterns over time, not demographic disparities. Use TrustyAI metrics for explainability and fairness.

Correct answer: B. InferenceService

The InferenceService CRD handles single-model serving definitions via KServe. ServedModel is used for multi-model serving runtimes, which serves as a quick way to eliminate the strongest distractor on the exam.

Correct answer: A. Secret

A Secret securely stores the credentials required to authenticate to external object storage. ConfigMaps lack encryption, making them unsuitable for sensitive data like AWS keys or S3 passwords.

Correct answer: A. Caikit + KServe

OpenShift AI uses Caikit with KServe by default to serve large language models efficiently. Seldon Core was the legacy runtime and is not the default high-performance choice for modern LLM workloads.

Correct answer: C. Secret

A Kubernetes Secret is recommended to store object storage credentials because it encrypts sensitive data at rest. PersistentVolumes are for long-term file storage, not credential management.

Correct answer: C. An individual script or notebook executed as a containerized step

Each node in an Elyra pipeline file maps to a specific operation, typically a script or notebook that runs as a containerized step. A physical cluster node is an infrastructure resource, not a pipeline execution step.

Correct answer: C. Data stored inside the mounted PVC directory is preserved

Data stored inside the mounted persistent volume claim directory remains intact during workbench pod restarts. Any data saved outside the PVC mount point is ephemeral and will be permanently lost when the container restarts.

Correct answer: D. A dedicated isolated Kubernetes namespace is provisioned

Creating a Data Science Project provisions a dedicated Kubernetes namespace to isolate resources and provide access controls. OpenShift AI maps these projects directly to namespaces for organizational and administrative boundaries.

Correct answer: D. The pipeline stops, marks that step as failed, and does not execute subsequent dependent steps

When a pipeline step fails, the execution stops and dependent downstream steps do not run. OpenShift AI Pipelines do not automatically reboot the cluster or blindly force final steps to run without their required upstream dependencies succeeding.

Correct answer: A. DataScienceCluster

The DataScienceCluster custom resource triggers the actual installation of OpenShift AI components. The DSCInitialization resource only handles prerequisite configurations, so it does not initiate the core component deployments needed for the platform.

Correct answer: A. Istio Service Mesh Framework

OpenShift Service Mesh uses Istio to handle traffic routing for single-model serving. OVNKubernetes handles the default pod networking, but it lacks the advanced routing rules needed for model serving traffic splitting and inference.

Correct answer: C. Node Feature Discovery and NVIDIA GPU Operator

Node Feature Discovery and the NVIDIA GPU Operator are the recommended tools for provisioning hardware accelerators. The Machine Config Operator handles general node configuration but lacks the specialized driver management required for functional GPU compute.

Correct answer: A. Kubeflow Pipelines architecture

Red Hat OpenShift AI pipelines are built on top of the Kubeflow Pipelines architecture. Argo CD is for continuous delivery, and Apache Airflow is a separate workflow tool neither integrated nor supported within the OpenShift AI dashboard environment.

Correct answer: D. Store credentials inside an OpenShift Secret referenced by the project

OpenShift Secrets securely isolate sensitive data like Git tokens, injecting them safely into the workbench environment without exposing them in code. Hardcoding credentials directly in notebooks or public files creates severe security vulnerabilities and fails the exam objectives.

Correct answer: C. Settings -> Notebook images section

The Settings menu contains the Notebook images section, which is the designated interface for managing custom notebook images and viewing their configuration statuses. Other sections like Pipelines or Deployed Models focus on execution logs and metrics, not administrative image settings.

Correct answer: C. managementState: Managed / Removed

Setting the managementState to Managed or Removed within the DataScienceCluster resource activates or deactivates specific OpenShift AI dashboard components. Other generic fields like enabled or status do not interface with the operator's component reconciler logic.

Correct answer: A. ModelMesh

ModelMesh is the target framework used by OpenShift AI to achieve high-density, multi-model serving on shared resources. KServe is a strong distractor, but ModelMesh specifically handles the intelligent caching and packing required for this density.

Correct answer: D. Model Serving Section

The Model Serving section of the dashboard reveals the live health of installed inference endpoints. Tabs for Data Science Pipelines handle execution graphs, while Settings manages cluster configurations, leaving Model Serving as the only correct monitoring view.

Correct answer: B. Endpoint URL

The endpoint URL parameter defines the specific network address used to connect with the storage provider when configuring a data connection. The bucket name specifies the container, but the endpoint dictates the exact network target.

Correct answer: B. /metrics

The /metrics endpoint is commonly queried to obtain Prometheus metrics from a served model. The /predict endpoint handles inference requests, and /healthz checks availability, but neither exposes the standard format required for cluster monitoring.

Correct answer: A. To provide an interactive containerized environment for running data science tools and code

The fundamental purpose of an OpenShift AI Workbench is to provide an interactive containerized environment for running data science tools and code. The distractors describe cluster operations like cron jobs and image builds, which are separate OpenShift platform features.

Correct answer: C. Model Quantization techniques

Model quantization techniques allow OpenShift AI to optimize and compress LLMs to fit onto smaller hardware footprints during serving operations. This process reduces the precision of the model weights, significantly lowering memory requirements without relying on generic container layering.

Correct answer: D. Red Hat OpenShift Serverless

Red Hat OpenShift Serverless handles the serverless scaling of models to zero when there is no traffic. While standard autoscalers manage load spikes, Serverless specifically enables scale-to-zero functionality through KServe and Knative Serving integrations.

Correct answer: B. Container image URL

Custom notebook images are imported as existing container images hosted in a registry, not uploaded archives. Look for the image pull URL format during the dashboard import workflow to identify the correct option easily.

Correct answer: C. Persistent Volume Claim

A Persistent Volume Claim provides stable block storage for workbenches, preserving your files across pod restarts. While S3 is supported for data connections, it is not automatically attached to the pod filesystem like the PVC.

Correct answer: C. Removed

Setting the management state to Removed in the DataScienceCluster resource instructs the operator to uninstall and delete that specific component. The Removed state actively reverses the installation, whereas other states simply pause reconciliation.

Correct answer: A. KServe

KServe supports scale-to-zero functionality for deployed models, making it ideal for cost-effective resource management. ModelMesh is designed for high-volume, always-on model fleets and does not scale down completely when idle.

Correct answer: C. Inspect the status fields of the DataScienceCluster custom resource to see if the KServe component is marked as Ready or Available

You verify component status by inspecting the DataScienceCluster custom resource using the CLI or web console. If KServe is enabled successfully, its status will display as Ready or Available.

Correct answer: B. Object storage via an S3-compatible API

OpenShift AI Pipelines require S3-compatible object storage to store artifacts and execution metadata. Block storage or ephemeral storage cannot be used because the pipeline engine needs S3 APIs.

Correct answer: D. To schedule a pipeline workflow to execute at specific periodic intervals

Configuring recurring runs allows you to schedule a pipeline to execute automatically at defined intervals. This automates continuous training workflows without requiring manual triggers.

Correct answer: C. Container image notebook stack selection

Selecting the notebook container image determines the base software stack and included Python packages for the workbench. PVC settings only affect storage capacity and mount paths, not the installed dependencies.

Correct answer: A. NVIDIA GPU Operator

The NVIDIA GPU Operator automates the management of GPU hardware components, including driver provisioning and monitoring. Other operators listed handle unrelated cluster functions like virtualization or network traffic routing.

Correct answer: D. TrustyAI Service instance

The TrustyAI Service instance must be deployed in a Data Science Project to intercept payload data and calculate fairness metrics. The other options are observability or deployment tools unrelated to bias monitoring.

Correct answer: B. Map the group to the rhods-users role

Mapping users to the rhods-users role grants standard access to create and manage Data Science Projects. Assigning cluster-admin or rhods-admins violates least privilege principles and poses severe security risks, making rhods-users the safe configuration.

Correct answer: C. Secret

Kubernetes Secrets are explicitly designed to securely store and inject sensitive data like passwords and API keys. Plain text is dangerously insecure, and ConfigMaps should only be used for non-confidential configuration data in your workbench environments.

Correct answer: C. Route

OpenShift Routes expose internal Services to generate secure, externally accessible URLs. A Service only provides internal cluster networking, so external traffic requires a Route to successfully reach your deployed machine learning inference models.

Correct answer: C. Red Hat OpenShift Cluster Monitoring (Prometheus/Grafana)

Red Hat OpenShift Cluster Monitoring uses Prometheus and Grafana by default. While you can integrate external tools, expect questions about default cluster metrics to target the built-in stack. External platforms require custom integration.

Correct answer: C. Remove the DataScienceCluster CR, delete the operator subscription, and clean up the component namespaces

Removing the DataScienceCluster custom resource, deleting the operator subscription, and cleaning namespaces correctly uninstalls the operator. Forcing API deletions or editing core infrastructure manifests is dangerous and incorrect.

Correct answer: B. OpenShift AI Dashboard

Data scientists use the OpenShift AI Dashboard to launch notebooks and manage models easily. While administrators rely on the OpenShift Web Console for platform management, the OpenShift AI Dashboard is the primary interface for users.

Correct answer: A. oc

The oc command-line tool is the primary interface for deploying custom resources in OpenShift AI. Tools like rhoai-cli are not standard deployment utilities. Memorize basic oc apply commands for custom resource management.

Correct answer: C. Deployed Models / Model Serving section

The Model Serving section displays active endpoints, response times, and request volumes for deployed models. The Pipelines tab focuses on execution workflows, not runtime inference metrics. Always check deployed models for endpoint health.

Correct answer: C. Red Hat OpenShift Service Mesh Operator

KServe relies on the Red Hat OpenShift Service Mesh Operator to handle structural networking and traffic management. Standard Kubernetes ingress lacks the advanced routing capabilities required, making the service mesh mandatory for model serving.

Correct answer: A. The pipeline run fails during validation or server initialization

Data Science Pipelines require an object storage bucket to store intermediate data artifacts. Without a configured connection, the pipeline run fails during validation or server initialization because it cannot persist the step outputs.

Correct answer: D. v2 Data Plane HTTP/gRPC API

The KServe v2 Data Plane HTTP or gRPC protocol is the industry standard for model inference requests. Exam takers should focus on v2 endpoints rather than custom REST wrappers when testing deployments.

Correct answer: C. The operator installs the component but stops updating or managing its state

Setting a component to Unmanaged installs it but halts operator reconciliation and lifecycle updates. This prevents accidental overwrites when administrators need to debug components manually.

Correct answer: C. OpenShift Serverless and OpenShift Service Mesh

KServe strictly requires OpenShift Serverless and Service Mesh to handle traffic routing. You must verify these operators are healthy before attempting to deploy single-model serving runtimes.

Correct answer: A. Pull Secret (kubernetes.io/dockerconfigjson)

A Kubernetes dockerconfigjson pull secret authenticates requests to private registries. You must properly link this secret to your service account to successfully pull custom images.

Correct answer: A. Retrieval-Augmented Generation (RAG)

Retrieval-Augmented Generation pulls outside documents into the model context window at inference time. Fine-tuning alters model weights, which defeats the requirement of avoiding a complete model retraining cycle.

Correct answer: D. A dedicated Red Hat OpenShift namespace

A Data Science Project maps directly to a dedicated Kubernetes namespace within OpenShift. This ensures workbenches and resources remain isolated, unlike cluster-wide definitions that lack multi-tenant boundaries.

Correct answer: C. DataScienceCluster

The DataScienceCluster custom resource initializes OpenShift AI components after operator installation. You edit this resource to toggle features, avoiding fictional resources like RHOAIManager or ClusterInitialization.

Correct answer: A. To configure GPU or specialized hardware access for notebook pods

Accelerator Profiles define how specialized hardware like GPUs are requested and mapped to notebook pods. They abstract hardware complexities, unlike distractors focused on networking or container image downloads.

Correct answer: A. AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY fields

AWS S3 data connections rely on standard AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY fields to authenticate properly. OpenShift AI expects these specific key names to mount your S3 storage automatically, unlike generic SSH or JSON deployment configurations.

Correct answer: C. OpenVINO Model Server runtime

The OpenVINO Model Server is pre-configured to serve predictive ONNX and OpenVINO models. It is the default supported runtime for high-performance inference in OpenShift AI, whereas standard frameworks lack this integrated support.

Correct answer: B. Culling configuration

Culling configuration determines when idle workbenches or notebooks are automatically stopped to free resources. Standard resource quotas merely limit usage, while pod disruption budgets only protect availability during voluntary disruptions.

Correct answer: A. Runtime Images configurations

Runtime Images configurations in the Elyra UI allow data scientists to specify custom environments for pipeline nodes. Component catalogs define the actual pipeline tasks, whereas runtime images dictate where those tasks execute.

Correct answer: B. nodeSelector

The nodeSelector parameter identifies the correct hardware nodes by matching defined labels. Tolerations merely allow scheduling onto tainted nodes, but they do not actively target specific instances like a node selector does.

Correct answer: B. ReadWriteOnce (RWO)

Personal storage disks in OpenShift AI use the ReadWriteOnce access mode by default. Since the volume is mounted to a single user's workbench pod, ReadWriteMany is unnecessary unless a shared file system is explicitly configured.

Correct answer: A. By applying an OpenShift LimitRange or ResourceQuota to the project namespace

Applying a LimitRange or ResourceQuota to the Data Science Project namespace enforces resource limits because the project maps directly to a namespace. Avoid modifying global node selectors or cluster roles for project-specific compute constraints.

Correct answer: C. Changing the update channel preference in the Operator Subscription manifest

Changing the update channel in the Operator Subscription manifest securely triggers OLM to handle the minor version upgrade. Manual yum upgrades or completely deleting the operator bypass OLM reconciliation and are not supported.

Correct answer: B. An admission webhook controlled by the RHOAI operator framework

An admission webhook intercepts workbench pod creation to inject Data Connection variables as environment variables. Manual copy-pasting or cron jobs would be unmanageable for automated pod scaling and rebuilds.

Correct answer: B. Grafana Dashboard Service

TrustyAI integrates directly with a Grafana Dashboard Service to visually display fairness metrics and drift trends. Tekton Results is for CI/CD pipelines and does not track model fairness metrics.

Correct answer: D. Knative Serving Engine

KServe relies on the Knative Serving engine to handle request routing and scale model replicas. Knative Serving provides the necessary capabilities to scale down to zero and manage incoming network traffic.

Correct answer: D. JupyterLab

JupyterLab is the primary integrated development environment provided out-of-the-box within OpenShift AI workbenches. It offers an interactive web interface highly optimized for data science and Python development.

Correct answer: D. Kubeflow Pipelines (KFP) SDK

Data Science Pipelines version 2 relies on the Kubeflow Pipelines SDK to author and compile workflow definitions. You compile your Python code into a YAML file that the pipeline engine can execute.

Correct answer: A. The underlying OpenShift namespace and all its resources are deleted

Deleting a Data Science Project removes the underlying OpenShift namespace and all associated resources. This destructive action means workbenches, pipelines, and configurations are permanently erased, making the hide-from-UI option incorrect.

Correct answer: C. An image pull secret containing valid registry credentials

An image pull secret containing valid registry credentials must be added to the namespace to access private workbench images. OpenShift uses this secret to authenticate container registry pulls securely.

Correct answer: A. TrustyAI

TrustyAI is the embedded service in OpenShift AI designed to monitor deployed models for bias and drift. While Prometheus monitors cluster metrics, TrustyAI directly handles explainability and fairness analysis.

Correct answer: B. S3-compatible Object Storage credentials

A Data Connection in OpenShift AI defines credentials for S3-compatible object storage endpoints. This configuration securely stores AWS-style access keys to allow workbenches and pipelines to retrieve data.

Correct answer: C. Data Science Project

Data Science Projects enable logical separation of resources like workbenches, pipelines, and models for different teams. They map directly to Kubernetes namespaces but provide specialized AI role-based access control.

Correct answer: D. DataScienceCluster

The DataScienceCluster custom resource defines the cluster-wide configuration for OpenShift AI components. Exam takers should look for the DSC configuration when managing platform features or installing operators across a distributed environment.

Correct answer: C. S3-compatible Object Storage

Data Science Pipelines relies on S3-compatible object storage to save artifacts like trained models. Storing files in ephemeral memory or Git repositories is impractical for binary model weights due to size limits and lack of accessibility.

Correct answer: C. Data Science Pipelines (based on Kubeflow Pipelines v2)

OpenShift AI Data Science Pipelines uses an orchestration engine based on Argo Workflows. Jenkins and GitLab are generic CI tools that lack native Kubernetes data science integrations required for this specific platform.

Correct answer: D. Persistent Volume Claims attached to /opt/app-root/src

Persistent Volume Claims attached to the workbench directory ensure user data persists across pod restarts. Ephemeral container memory is wiped when the pod shuts down, making it completely unsuitable for saving work.

Correct answer: C. DataScienceCluster

The DataScienceCluster custom resource definition manages global platform configurations and components in Red Hat OpenShift AI. DSCInitialization handles initial setup, but the DataScienceCluster resource controls the active global cluster configuration.

Correct answer: A. Git Extension Interface Panel

The JupyterLab Git Extension Interface Panel provides a graphical way to manage version control collaboratively. While terminal commands are available, the interface panel is the integrated solution designed for this workflow.

Correct answer: D. datasciencecluster.opendatahub.io

The DataScienceCluster custom resource definition belongs to the datasciencecluster.opendatahub.io API group. OpenShift AI builds upon OpenDataHub foundations, so its core configurations share this API lineage rather than generic Red Hat namespaces.

Correct answer: A. The storage remains preserved and reattaches when the workbench is started again

Persistent volume claims retain data when a workbench stops. The storage reattaches upon restarting, ensuring notebooks and files are not lost. Data is only removed if the persistent volume claim itself is explicitly deleted.

Correct answer: D. POST

Model inference testing requires sending a data payload to the endpoint, making POST the correct method. GET retrieves information without a body, so it cannot deliver the required payload for inference.

Correct answer: D. Create an inference service token or configure ServiceAccount authentication tokens

Creating an inference service token configures the required authentication for external clients. Disabling network policies or allowing anonymous access violates basic security practices and will not securely grant valid access.

Correct answer: D. YAML or Python decorators defining the component container specs

Kubeflow Pipelines rely on component definitions formatted as YAML or Python decorators. These formats specify the container image and inputs, which is required for the pipeline execution engine.

Correct answer: A. ReadWriteMany (RWX)

ReadWriteMany allows multiple nodes to mount the volume simultaneously for reading and writing. ReadWriteOnce restricts access to a single node, preventing concurrent use by multiple data scientists.

Correct answer: A. RHOAI Operator

The Red Hat OpenShift AI Operator handles installation and lifecycle management using a custom resource. Knowing the operator's role is crucial for administering the platform and troubleshooting component deployment.

Correct answer: A. Ingress Gateway

The OpenShift Service Mesh uses an Ingress Gateway to route external HTTP traffic securely to internal services like models. Standard cluster services such as NodePort or ClusterIP do not provide the required secure external routing by themselves.

Correct answer: A. edit

Assigning the edit role in a Data Science Project allows a user to modify workbenches without granting project deletion privileges. The admin role grants excessive permissions, while viewer strictly limits read access.

Correct answer: A. Tolerations and Node Selectors parameters

Tolerations and node selectors control where the OpenShift AI notebook pods land in the cluster. The storage and ingress options fail because they manage capacity and routing, not the physical scheduling of the compute workload.

Correct answer: A. The individual step execution logs available inside the Pipeline Run graph view

The Pipeline Run graph view provides direct access to the individual execution logs for each failed step. The infrastructure event viewer or operator history will not show the specific data science code errors needed to troubleshoot the pipeline failure.

Correct answer: B. Prometheus

Prometheus is the underlying metrics storage layer used by OpenShift to capture data from served models. The other options are incorrect because Jaeger handles tracing, while Elasticsearch and Thanos serve different data retrieval purposes.

Correct answer: C. DashboardConfig

The DashboardConfig custom resource definition allows administrators to manage cluster-wide settings for the OpenShift AI dashboard. It controls dashboard features and namespace futures, while the DataScienceCluster resource manages the core components.

Correct answer: A. Tekton (OpenShift Pipelines)

OpenShift AI relies on Tekton, provided by OpenShift Pipelines, to execute Data Science Pipelines. Understanding the underlying Tekton engine is crucial for debugging pipeline step failures or checking pod logs during execution.

Correct answer: A. DataScienceCluster

The DataScienceCluster custom resource is the primary configuration object used to enable and disable OpenShift AI components cluster-wide. Remember to use oc get dsc to quickly verify your component installations.

Correct answer: D. Using the integrated JupyterLab Git extension or terminal

Within an OpenShift AI Workbench, users clone repositories directly via the built-in JupyterLab Git extension or terminal. This provides a secure and native way to pull code into the active persistent workspace.

Correct answer: C. Knative Pod Autoscaler

The Knative Pod Autoscaler handles rapid scaling for KServe based on immediate request traffic. The standard Kubernetes Horizontal Pod Autoscaler is slower and relies on aggregate CPU metrics rather than scaling to zero.

Correct answer: A. NVIDIA GPUs with the NVIDIA GPU Operator installed

NVIDIA GPUs, managed by the NVIDIA GPU Operator, are the primary supported hardware for running large generative models in OpenShift AI. CPUs and NVMe drives cannot handle the compute demands of training large models.

Correct answer: D. ModelMesh

ModelMesh is the sub-component that provides multi-model serving capabilities within OpenShift AI. KServe is often confused here, but KServe primarily handles single-model serving deployments, whereas ModelMesh explicitly manages high-throughput multi-model serving.

Correct answer: A. Disparate Impact or performance disparity metrics

Disparate impact or performance disparity metrics identify whether a model's accuracy varies significantly across diverse data subsets. TrustyAI specifically calculates these fairness metrics, whereas memory, network, and IOPS options represent infrastructure monitoring rather than model fairness.

Correct answer: B. vLLM or Caikit+TGIS serving runtimes

The vLLM or Caikit+TGIS serving runtimes are specialized runtimes ideal for efficiently serving Large Language Models in OpenShift AI. OpenVINO is generally used for traditional inference, while Jupyter notebooks are for development, not production serving.

Correct answer: A. OpenShift OAuth access tokens or ServiceAccount tokens

OpenShift AI uses OpenShift OAuth access tokens or ServiceAccount tokens to securely authenticate internal component requests. Basic HTTP credentials or cleartext headers are insecure and are not used by the platform.

Correct answer: B. Istio Ingress Gateway

The Istio Ingress Gateway manages external traffic routing for KServe single-model serving in OpenShift AI. While standard OpenShift Routes handle typical web traffic, KServe relies on Service Mesh components for advanced ingress control.

Correct answer: A. Concurrency target limit setting

The concurrency target limit setting specifies the maximum number of parallel requests a single model replica handles before autoscaling triggers. CPU quotas merely throttle processing power and do not strictly cap simultaneous inbound request counts.

Correct answer: D. Red Hat OpenShift Service Mesh

Red Hat OpenShift Service Mesh provides the Istio capabilities necessary for routing ModelMesh traffic. The 3Scale Operator handles external API management rather than the internal pod networking required for serving multiple models.

Correct answer: C. vLLM

The vLLM runtime is optimized for high-throughput serving of large language models within OpenShift AI. While ModelMesh serves standard predictive models, vLLM provides the specialized memory management required for massive LLM inference.

Correct answer: D. USER_NAMESPACE

The USER_NAMESPACE environment variable is automatically injected into OpenShift AI workbenches. It allows scripts and pods to dynamically reference the specific namespace where the logged-in user is operating.

Correct answer: C. gRPC / HTTP

KServe natively supports both gRPC and HTTP protocols for high-performance inference requests. gRPC is preferred for lower latency, while standard HTTP handles typical REST payloads.

Correct answer: A. oc

Administrators use the oc command-line tool to check the status of custom resources and operator deployments in OpenShift. There is no default rhoai command-line utility included for cluster inspection tasks.

Correct answer: D. Triton Inference Server

ModelMesh uses the Triton Inference Server runtime to serve TensorFlow models natively. TorchServe and vLLM cater to different framework priorities like large language models rather than general TensorFlow inference.

Correct answer: C. Elyra Pipeline Editor

The Elyra Pipeline Editor provides a visual interface for building, scheduling, and automating machine learning workflows directly within Jupyter. The Model Mesh Dashboard handles model deployment rather than pipeline creation.

Correct answer: B. Knative Pod Autoscaler (KPA)

Knative Pod Autoscaler automatically scales single-model serving runtimes based on incoming HTTP requests. Remember that while standard OpenShift uses HorizontalPodAutoscaler for CPU metrics, model serving routes rely specifically on KPA for rapid, request-driven scaling.

Correct answer: D. DataScienceCluster Spec Component Management

The DataScienceCluster custom resource manages component states globally. Administrators edit its spec to enable, disable, or manage component sizes, rather than altering node requests directly. Focus on managing custom resources for operator configurations.

Correct answer: C. Container Resource Requests / Limits

Container resource requests and limits define CPU and memory overhead for pipeline tasks before execution. This ensures pods have adequate resources. Elyra preferences and parameters do not directly control compute resource allocations.

Correct answer: D. Tekton Pipeline Task Pod logs viewable in Developer Console

OpenShift AI pipelines run as Tekton tasks, making the Developer Console pod logs the best place to find execution outputs. Operator diagnostics or node system logs will not show specific pipeline task failures clearly.

Correct answer: A. Vector/Fluentd via Cluster Logging Operator

The Cluster Logging Operator uses Vector or Fluentd to capture cluster logs seamlessly. OpenShift GitOps handles deployments, not log aggregation. Rely on built-in OpenShift logging operators to collect pipeline outputs.

Correct answer: A. By utilizing pipeline parameters within the dashboard run configuration interface

You configure parameters dynamically by utilizing pipeline parameters within the dashboard run configuration interface. Modifying manifests or global operator configs is incorrect because runtime values are set directly in the UI.

Correct answer: D. Accelerator Profiles

When creating a workbench in OpenShift AI, you use the Accelerator Profiles selection to allocate GPUs. While node selectors and tolerations exist underneath, the UI specifically requires choosing an accelerator profile.

Correct answer: A. OpenVINO Model Server

OpenVINO Model Server is the default and most common runtime used within OpenShift AI for deploying ONNX models. Other servers like Triton or TF Serving are supported but OpenVINO is the platform standard.

Correct answer: D. MLMD (Machine Learning Metadata)

MLMD, or Machine Learning Metadata, captures pipeline step metadata and execution history. Prometheus handles metrics while Grafana handles visualization, but MLMD specifically stores the artifact lineage for data science pipelines.

Correct answer: C. Red Hat OpenShift Data Foundation Operator

Red Hat OpenShift Data Foundation provides the underlying storage required for OpenShift AI. Be careful with ambiguity here, as the OpenShift Pipelines Operator is also technically required for data science pipelines functionality.

Correct answer: A. ModelMesh Serving Engine

ModelMesh serves as the default multi-model serving engine in OpenShift AI. The KServe engine requires OpenShift Serverless and Service Mesh, which makes it distinct from the default configuration.

Correct answer: B. Red Hat Certified Notebook Image

Red Hat Certified Notebook Images come pre-packaged with popular data science libraries. Using certified images ensures software compatibility and avoids dependency conflicts when launching workbenches.

Correct answer: A. Configure the workbench to select a container image flavor that includes GPU drivers and specify the requested GPU count

Selecting a GPU-compatible image and requesting the hardware count tells OpenShift AI to schedule the notebook on a GPU node. Avoid options suggesting manual PCIe mapping or kernel recompilation, as OpenShift manages hardware allocation automatically.

Correct answer: B. The model repository URI and an active Data Connection secret

Model serving requires the storage location and linked credentials to fetch the model files. OpenShift AI uses a Data Connection to provide these credentials securely, making manual tokens or raw binary strings unnecessary.

Correct answer: B. Data Drift

Data drift measures statistical changes between the training dataset and live production inputs. Mean Squared Error evaluates prediction accuracy, whereas bias coefficients check fairness rather than input distribution shifts.

Correct answer: B. OdhDashboard

The OpenShift AI dashboard provides the graphical interface for creating projects and workbenches. While the OpenShift Web Console manages cluster administration, it is not the specific tool for data science workflows.

Correct answer: B. ReadWriteMany (RWX)

ReadWriteMany allows the persistent volume to be mounted by multiple workbenches simultaneously. ReadWriteOnce restricts access to a single pod, preventing concurrent multi-user access required for shared storage.

Correct answer: A. The workbench pod remains in a Pending status state due to unschedulable resources

When a workbench requests resources that exceed what is available, the pod remains in a Pending state because the scheduler cannot find a suitable node. OpenShift will not silently fall back to regular CPU units when an explicit GPU allocation fails.

Correct answer: A. Prometheus and Grafana via the NVIDIA GPU Operator dashboards

The NVIDIA GPU Operator integrates with Prometheus and Grafana to provide dashboards for GPU metrics, memory, and temperature. The OpenShift AI dashboard manages pipelines, but it does not natively monitor hardware metrics.

Correct answer: A. ContainerSize

The ContainerSize parameter directly defines the CPU and memory resource limits allocated to a workbench container. Resource quotas and limit ranges apply broad constraints across a namespace rather than defining specific workbench sizes.

Correct answer: C. The specific pipeline step pod is terminated with an OOMKilled error

The pod running the pipeline step gets terminated with an OOMKilled error when it breaches its configured memory limits. Kubernetes enforces these hard limits, so the pod cannot ignore them or automatically scale the underlying worker node CPU.

Correct answer: C. Node Feature Discovery (NFD) Operator

The Node Feature Discovery Operator labels nodes with hardware accelerators like NVIDIA GPUs. The OpenShift AI Operator relies on these labels to schedule workbench or inference pods onto the correct compute nodes.

Correct answer: C. OpenShift AI Dashboard

The OpenShift AI Dashboard serves as the primary interface for launching Jupyter notebooks and data science workbenches. Standard OpenShift web consoles do not provide these specialized machine learning interfaces by default.

Correct answer: B. Tekton PipelineRun YAML or Tekton format

The Elyra pipeline editor exports visual workflows into Tekton PipelineRun YAML formats for execution. Because OpenShift AI Pipelines natively rely on Tekton, Argo or Jenkins formats will not execute natively.

Correct answer: C. Elyra Pipeline Editor

The Elyra Pipeline Editor is the visual tool integrated into Jupyter workbenches for building and submitting pipelines. It is explicitly designed to generate the required Tekton YAML for the underlying pipeline engine.

Correct answer: C. Data Science Cluster Controller Manager pod

The Data Science Cluster Controller Manager pod handles the reconciliation of the DataScienceCluster custom resource. This controller ensures the requested OpenShift AI components are properly deployed and maintained.

Correct answer: B. oc get csv -n redhat-ods-operator

Checking the ClusterServiceVersion or CSV confirms the operator was installed successfully via OperatorHub. This command displays the installation phase and verifies the operator is resolving correctly.

Correct answer: B. OpenShift OAuth Proxy

The OpenShift OAuth Proxy acts as the authentication wrapper securing access to workbench instances. It integrates directly with OpenShift OAuth to ensure only authorized users can reach the notebooks.

Correct answer: B. OpenShift OAuth Proxy

The OpenShift OAuth Proxy handles user authentication and session redirection when accessing workbenches. Other identity brokers like Keycloak are distractors not directly integrated into dashboard launches.

Correct answer: C. Notebook

The Notebook custom resource manages workbench instances for data scientists. While Workbench sounds correct, the underlying Kubernetes object used by the operator is the Notebook CR.

Correct answer: C. To act as an isolated Kubernetes namespace with managed access controls and project-specific resources

A Data Science Project acts as an isolated Kubernetes namespace with managed access controls. It organizes resources like workbenches and pipelines rather than handling CI/CD builds or physical node allocation.

Correct answer: C. Node Feature Discovery (NFD)

Node Feature Discovery detects specific hardware accelerators on nodes and applies the appropriate labels. The NVIDIA GPU Operator uses these labels, but NFD is the service that actually discovers and applies them.

Correct answer: D. Kubeflow Pipelines SDK

The Kubeflow Pipelines SDK is the primary framework for creating reusable components in OpenShift AI data science pipelines. Ansible or AMQ Streams handle unrelated automation tasks.

Correct answer: B. JupyterLab IDE Extension

Elyra operates as a visual builder extension directly inside the JupyterLab IDE. The OpenShift console manages cluster administration, not visual pipeline construction.

Correct answer: B. Settings -> Accelerator profiles

Administrators manage accelerator profiles under Settings and then Accelerator profiles. Hardware profiles represent a common distractor but are not used for GPU configuration.

Correct answer: A. Project Admin role permissions

Users need Project Admin permissions to manage group access within their specific Data Science Project. Viewers and standard cluster operators lack delegation capabilities.

Correct answer: C. Creating custom notebook container images containing needed libraries pre-baked

Creating custom notebook images with required libraries pre-installed ensures packages are securely available in disconnected environments. Avoid direct internet dependencies because restricted networks block direct downloads and external repository access.

Correct answer: C. Vector Databases like Milvus, Chroma, or Redis

Vector databases such as Milvus or Chroma are commonly integrated with OpenShift AI to handle embeddings and vector searches for RAG. Standard relational databases or simple persistent volumes lack the optimized indexing needed for fast similarity searches.

Correct answer: C. Node Affinity Rules

Node affinity rules are the standard mechanism to target specific hardware types, ensuring workbenches only run on nodes with NVIDIA accelerators. Taints and tolerations allow pods to tolerate conditions but do not exclusively guarantee hardware scheduling.

Correct answer: D. NVIDIA GPU Operator with MIG profiles

The NVIDIA GPU Operator manages Multi-Instance Graphics configurations, allocating specific GPU slices to different workbenches. Standard OpenShift route rules or node selectors lack the hardware-level partitioning needed for MIG profiles.

Correct answer: B. DashboardConfig

The DashboardConfig custom resource manages dashboard settings including custom workbench images. OpenShift AI relies on standard operators like ODH DashboardConfig rather than generic custom resource definitions to control user interfaces.

Correct answer: A. Red Hat OpenShift Data Foundation (ODF)

Red Hat OpenShift Data Foundation provides seamless persistent storage and is heavily used for local object store simulations in OpenShift AI. Ephemeral container volumes and hostpath providers do not provide reliable persistent storage.

Correct answer: D. Output Files configuration tracking paths

Elyra uses Output Files configuration tracking paths to define where generated files are exported for secure retrieval after execution. Git push targets handle source code versioning, not pipeline output artifacts.

Correct answer: D. OpenVINO IR (XML/BIN) or ONNX format

OpenVINO serving runtimes generally expect models in the OpenVINO IR XML/BIN format or ONNX format for optimized inference. Pickle files are unsafe and specific to Python objects, whereas Protocol Buffers are TensorFlow-specific.

Correct answer: B. Kubeflow

Kubeflow is the correct answer because OpenShift AI builds upon its concepts and components. While MLflow is heavily integrated, Kubeflow serves as the foundational upstream project for the pipeline architecture.

Correct answer: A. Kubernetes Secrets

Data connections in OpenShift AI are stored as Kubernetes Secrets, which are then mounted as environment variables in the notebook pods. ConfigMaps lack built-in security for sensitive data like passwords and access keys.

Correct answer: C. ODS_DATA_CONNECTION

The dashboard injects data connection details into custom workbenches using the ODS_DATA_CONNECTION prefix. Recognizing this variable is vital when configuring custom notebook environments to read S3 storage details programmatically.

Correct answer: B. ReadWriteMany

ReadWriteMany is required for shareable JupyterLab workspaces so multiple pods can mount the volume simultaneously. ReadWriteOnce restricts access to a single node, which prevents proper multi-user collaboration.

Correct answer: A. Amazon S3 or compatible object storage

Data Science Pipelines use Amazon S3 or compatible object storage to store run artifacts and compiled workflow files. Local or ephemeral storage options lack the persistence necessary for reliable pipeline execution across nodes.

Correct answer: A. oc logs deployment/rhods-operator -n redhat-ods-operator

The oc logs command targeting the rhods-operator deployment retrieves the runtime status and errors. The other options are invalid because they reference nonexistent system flags or operator names that do not directly expose the required troubleshooting logs.

Correct answer: A. Phase: Ready

The DataScienceCluster custom resource uses the Ready phase to indicate that a component is fully installed and operational. When checking component status, look specifically for the phase condition rather than just a generic installed or ready string.

Correct answer: C. cluster-admin

Configuring global image streams and managing custom notebook images for all users requires cluster-wide administrative privileges. The cluster-admin role provides the necessary access to modify resources across all namespaces.

Correct answer: C. Pipeline Scheduler (integrated with KFP)

OpenShift AI uses the native Pipeline Scheduler integrated with Kubeflow Pipelines to trigger recurring runs at specific intervals. This avoids needing standalone Kubernetes CronJobs for standard automated retraining tasks.

Correct answer: B. It scrapes metrics from user workloads and surfaces them to Prometheus

The Smart Gateway collects metrics from user workloads and sends them to Prometheus for monitoring. While the Kubernetes Horizontal Pod Autoscaler uses these metrics to scale pods, it only reacts to the data collected by this component.

Correct answer: B. Structured JSON objects matching the inference API schema

Interactions with models served by vLLM use structured JSON objects matching the inference API schema. Plain text or CSV formats lack the required structure to define model parameters and prompt templates effectively.

Correct answer: B. Cluster Autoscaler Engine

The Cluster Autoscaler adds or removes compute nodes when pods fail scheduling due to insufficient hardware. Node Feature Discovery labels existing hardware but does not provision new capacity.

Correct answer: B. Red Hat OpenShift Data Foundation

OpenShift Data Foundation provides the reliable shared storage needed for collaborative data science workspaces. HostPath or cloud-specific block storage lack the RWX capabilities needed for simultaneous notebook access.

Correct answer: C. StorageClass

A StorageClass defines the storage capacity and performance characteristics available for a workspace by provisioning specific volume types. ResourceQuota limits resource consumption, while NetworkPolicy handles routing and isolation rules rather than underlying storage capabilities.

Correct answer: C. ImageStream

Administrators register custom notebook images by creating standard OpenShift ImageStreams in the dedicated namespace. The dashboard reads these ImageStreams to populate the available image list during workbench creation.

Correct answer: C. Kubeflow Pipelines

OpenShift AI relies on Data Science Pipelines, which are built on top of Argo Workflows, to manage and execute directed acyclic graphs. While Argo handles the underlying execution, the pipeline engine is managed by the pipeline component.

Correct answer: C. gRPC

gRPC provides the low-latency, high-throughput communication often used for model inference payloads compared to standard HTTP. While frameworks support REST, the binary serialization of gRPC reduces network overhead for heavy payloads.

Correct answer: A. Worker Nodes with GPU/Accelerators

Machine learning training tasks are dispatched to worker nodes configured with accelerators like GPUs. On the exam, remember that infrastructure nodes handle routing and logging, while control plane nodes strictly manage cluster state.

Correct answer: D. KServe ServingRuntime nodeSelector / Toleration mappings

You direct models to specialized hardware by applying nodeSelector and Toleration mappings within the KServe ServingRuntime configuration. Load balancing and PodDisruption budgets manage availability and traffic, but they cannot enforce hardware targeting.

Correct answer: D. redhat-ods-operator

The redhat-ods-operator namespace houses the core controllers and operands required to run OpenShift AI. While standard operators often sit in openshift-operators, Red Hat OpenShift AI isolates its specific controllers in this dedicated namespace.

Correct answer: A. OpenVINO Model Server (OVMS)

OpenVINO Model Server is specifically optimized for deploying Intel AI models with high performance. Triton or TensorFlow Serving serve broader model types but lack the native Intel hardware optimizations expected here.

Correct answer: C. Notebook Controller ConfigMap

The Notebook Controller ConfigMap applies cluster-wide environment variables to all Jupyter notebooks. The OdhDashboard Custom Resource configures UI settings but does not inject cluster-wide notebook environments.

Correct answer: B. Elyra Pipelines

Elyra pipelines allow direct export and execution of notebook code as reproducible pipeline steps. Tekton triggers handle automated continuous deployment tasks instead of translating interactive notebook logic into executable workflows.

Correct answer: C. Managed

Setting the component status to Managed instructs the operator to install and reconcile the specified component. Options like Removed delete the component, while Enabled and active are not valid states within the DataScienceCluster custom resource specification.

Correct answer: A. id_rsa / ssh-privatekey config

OpenShift AI workbenches look specifically for the ssh-privatekey key within a Secret to clone repositories securely. OAuth tokens or standard passwords are used for alternative Git authentication methods, not interactive SSH key operations.

Correct answer: C. rhods-dashboard-

The rhods-dashboard pod routes web traffic for the OpenShift AI user interface. Knowing the specific pod names helps quickly isolate application logs during troubleshooting tasks.

Correct answer: C. .pipeline JSON file

Elyra exports pipeline definitions as a .pipeline JSON file for execution on OpenShift Pipelines. Python scripts and tar archives are runtime artifacts, not the export format needed to launch the workflow.

Correct answer: D. Tekton backend

OpenShift AI pipelines are built on the Tekton backend. The question is technically outdated because current exams use the Data Science Pipelines Application, though Tekton remains the underlying execution engine.

Correct answer: D. Knative Serving component architecture

Knative Serving handles scale-to-zero functionality for KServe model deployments. Standard Horizontal Pod Autoscalers cannot scale to zero without an external metric scaler, making Knative the required component.

Correct answer: B. Environment Variables from Secrets

Attaching a data connection to a workbench injects the associated Kubernetes Secret as environment variables. This mechanism ensures your notebook code can dynamically reference cloud storage credentials without hardcoding them.

Correct answer: C. YAML or JSON

Kubeflow Pipelines SDK compiles pipelines into YAML or JSON format. Remember that compiled pipelines are stored as compressed YAML files, making options like XML or TOML incorrect for OpenShift AI pipeline definitions.

Correct answer: D. Sharing single serving runtime resources among multiple model files for high density efficiency

Multi-model serving maximizes resource density by sharing a single serving runtime among multiple deployed models. Single-model serving isolates compute per model, making it less efficient for large-scale deployments.

Correct answer: B. In an S3-compatible object storage database or database custom resources

TrustyAI stores telemetry data in an S3-compatible object storage bucket configured during installation. Local memory or node journals are not used because persistent metrics are required.

Correct answer: C. PipelineRun

The PipelineRun custom resource tracks the execution history and status of a triggered pipeline. TaskRun only tracks individual tasks, whereas PipelineRun represents the entire workflow execution.

Correct answer: C. By intercepting traffic via a payload logging feature enabled on the model mesh or KServe runtime

TrustyAI captures payloads by intercepting traffic through the payload logging feature enabled on the runtime. Dynamic source code modification is incorrect because interceptors operate externally.

Correct answer: B. ServiceMesh Ingress Gateway Route

A ServiceMesh Ingress Gateway Route securely routes external client traffic to the deployed model. ClusterIP and NodePort services fail to provide the required secure external routing.

Correct answer: B. /opt/app-root/src

OpenShift AI workbenches default to the /opt/app-root/src directory for storing user files. The /home/jovyan path is common in vanilla Jupyter deployments but fails on OpenShift.

Correct answer: B. Git Repository (GitHub, GitLab, Gitea)

A Git repository serves as the ultimate source of truth for declarative infrastructure and configurations in GitOps workflows. Container registries only store images and lack the desired state definitions required for continuous reconciliation.

Correct answer: D. OpenShift Route via Istio Ingress

KServe creates an OpenShift Route through the Istio Ingress Gateway to expose inference endpoints externally. Edge routes are typically used for standard cluster traffic, but Service Mesh integration dictates the Istio requirement here.

Correct answer: C. aws-storage-bucket-name

The aws-storage-bucket-name field specifies the S3 bucket location inside an OpenShift AI data connection Secret. Knowing these exact keys is critical for manually creating or verifying data connections via the command line.

Correct answer: D. Workbenches must be stopped, edited in the dashboard settings configuration, and restarted

You must stop the workbench, change the container configuration in the dashboard, and restart it to apply new memory limits. Live patching custom resources is not supported, making the manual restart process the expected operational procedure on exams.

Correct answer: A. Kubernetes

Red Hat OpenShift is built on top of Kubernetes, which acts as the foundational orchestration engine for managing container workloads. This baseline knowledge is necessary for navigating cluster resources during the exam.

Correct answer: C. JupyterLab Git Extension

The JupyterLab Git Extension provides native version control integration within the workbench interface. OpenShift GitOps operates at the cluster level and does not directly embed into the interactive notebook environment.

Correct answer: B. The pipeline file properties metadata configuration parameters

Elyra stores external repository dependencies and configurations directly within the pipeline file properties metadata. Do not confuse this with the standard Git config, which only handles local source control mappings rather than visual pipeline execution dependencies.

Correct answer: B. Tekton PipelineRun Spec Retry

Data Science Pipelines in OpenShift AI use Tekton under the hood, so retries are configured in the PipelineRun spec. Argo might look familiar, but the current native pipeline engine is Tekton.

Correct answer: D. restricted-v2 or customized default profiles

Restricted or carefully customized SCC profiles enforce least privilege, keeping workbench containers secure. Privileged or anyuid access defeats these restrictions and is rarely granted to standard users.

Correct answer: B. ServingRuntime CRD instance

A ServingRuntime custom resource instance defines the runtime environment and infrastructure required by ModelMesh to host models. This object tells the controller how to deploy the necessary model servers.

Correct answer: B. DSCInitialization

The DSCInitialization custom resource manages global system configurations required before deploying components. This configuration must be applied successfully so the DataScienceCluster CR can properly initialize the operator.

Correct answer: D. Worker nodes equipped with NVIDIA GPUs

Worker nodes equipped with NVIDIA GPUs provide the specialized hardware required to run heavy LLM training workloads efficiently. Master and infrastructure nodes handle cluster operations and cannot handle intensive training.

Correct answer: A. PersistentVolumeClaim

A PersistentVolumeClaim requests a logical segment of storage and mounts it into a single workbench pod. Expect PVCs whenever a question asks about persistent storage allocation for notebooks.

Correct answer: D. Object Storage (S3 Compatible)

Data Science Pipelines require an S3-compatible object storage bucket to store artifacts. File or block storage options are incorrect because the pipeline engine specifically interfaces with object storage APIs.

Correct answer: A. To hold the default workbench image streams and components

The designated namespace is used to store the default workbench image streams and required components. Knowing where custom images live is an essential administrative exam cue.

Correct answer: C. KServe v2 Data Plane API Schema

Triton integrates with OpenShift AI by exposing the standard KServe v2 Data Plane API schema. Remember that legacy custom protocols fail modern multi-model server compatibility checks.

Correct answer: B. oc delete project

The correct answer is oc delete project because a Data Science Project acts as an OpenShift namespace. Use the standard command line interface tool to remove the namespace entirely from the cluster.

Correct answer: C. oc get crd | grep -iE 'smarter|kafk|rhods|datascience|kserve|trustyai'

Using standard OpenShift commands to list and filter CRDs is a valid troubleshooting method. While oc get crds is the base requirement, the specific grep filters shown here help identify OpenShift AI components quickly.

Correct answer: D. Active or Ready

A status of Active or Ready indicates the project is fully provisioned and prepared for workloads. Pending means resources are still allocating, while Terminating indicates the project is actively being destroyed.

Correct answer: A. Red Hat OpenShift AI does not store user data in a dedicated external DB; it uses cluster resources.

OpenShift AI does not store user data in a dedicated external database, instead relying on internal cluster resources and CRDs. Options like PostgreSQL or MongoDB are distractors not used for dashboard configurations.

Correct answer: B. The DataScienceCluster Custom Resource spec section

The DataScienceCluster custom resource controls which components are enabled or disabled in OpenShift AI. Expect to edit this resource directly when you need to toggle specific platform features on or off during or after deployment.

Correct answer: A. Code version control, sharing, and tracking adjustments over time

Integrating Git with workbenches provides standard version control so teams can share code and track changes over time. While Git is excellent for scripts, remember to use cloud storage or data connections for large binary model weights.

Correct answer: C. Workbench

A workbench provides the interactive Jupyter environment needed for data scientists to write and execute code. Remember that data connections handle storage and model servers handle deployments.

Correct answer: D. DSCInitialization

The DSCInitialization custom resource must be configured first because it sets up global settings and prerequisites. The DataScienceCluster CR relies on this initialization to properly deploy and manage the individual components.

Correct answer: A. Prometheus and Grafana dashboards stack

OpenShift AI relies on the native OpenShift monitoring stack powered by Prometheus and Grafana to observe system health. Grafana dashboards provide the visual interface to track serving performance metrics directly.

Correct answer: B. kfp Python software development kit library package

The Kubeflow Pipelines SDK is the standard method for defining and submitting OpenShift AI pipelines. Although you manage infrastructure with OpenShift tools, data scientists specifically rely on the Python kfp package for pipeline code.

Correct answer: D. oc logs command line terminal

The oc logs command retrieves direct standard output and error streams from specific operator pods. While distributed tracing is great for network latency, oc logs is your primary tool for immediate operator initialization failures.

Correct answer: D. Workbenches interface environment running within a project space

Workbenches provide the integrated development environments, like JupyterLab, used by data scientists within an OpenShift AI project. Pipelines and model serving handle execution and deployment rather than interactive development.

Correct answer: B. To discover, allocate, and configure hardware accelerators like NVIDIA GPUs for workbenches and serving

Accelerator Profiles manage the discovery, allocation, and configuration of hardware accelerators like GPUs. This feature specifically targets workbenches and model serving rather than software downloads or build node scaling.

Correct answer: A. Configure an Image Stream or add it in the cluster settings

Custom workbench images are added by configuring an Image Stream or using the cluster settings. Data connections configure external storage rather than dashboard images, making the cluster settings approach correct.

Correct answer: D. It allows users to view project configurations, workbenches, and pipelines without making changes

The rhoai-viewers role grants read-only access to OpenShift AI project resources like configurations and pipelines. This prevents users from modifying resources or deleting deployments, unlike administrative roles.

Correct answer: D. Packing multiple models efficiently into shared memory and containers

ModelMesh efficiently packs multiple models into shared memory and containers to maximize density. It focuses on runtime inference rather than building containers or scheduling persistent storage backups.

Correct answer: A. REST / gRPC Interface

The KServe v2 protocol uses standard REST and gRPC interfaces for model inferencing. OpenShift AI leverages this interface directly, avoiding legacy formats like SOAP XML or unrelated channels like GraphQL.

Correct answer: D. Selected Workbench Notebook Container Image

The selected Workbench Notebook Container Image dictates the base software packages, Python libraries, and IDE tools available. Compute profiles and PVC storage classes only allocate hardware resources rather than defining the software environment.

Correct answer: C. Pre-installing specific company libraries, tools, and configurations

Building a custom notebook image allows teams to pre-install specific corporate libraries, tools, and configurations. This saves developers time and ensures standardization across the organization without requiring manual setup.

Correct answer: D. Red Hat OpenShift AI Operator

You must install the Red Hat OpenShift AI Operator to provision the core components and manage the lifecycle of the OpenShift AI instance. Operators like GitOps are complementary but not required for the initial system deployment.

Correct answer: A. The data is permanently lost because the container filesystem is ephemeral

The correct answer is correct because container file systems are ephemeral by design, meaning any data not saved to a mounted persistent volume is destroyed when the pod restarts. Remember that workbenches only preserve data explicitly written to the mounted persistent storage folder.

Correct answer: C. The serving runtime fails to initialize or reload the model

The correct answer is correct because serving runtimes require the data connection secret to authenticate and pull the model artifacts from storage. Without the secret, the runtime cannot fetch the model and will fail during initialization.

Correct answer: D. To store pipeline artifacts, metrics, and execution logs

The correct answer is correct because the pipeline server uses object storage to persist run artifacts, metrics, and logs generated during workflow execution. This ensures data is retained and accessible even after the pipeline pods terminate.

Correct answer: C. AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY

OpenShift AI uses standard AWS environment variables like AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY to pass S3 credentials into the workbench. This allows default boto3 and other compatible SDKs to authenticate automatically without manual code changes.

Correct answer: C. Persistent Volume Claim size specified during workbench creation

The workspace storage size is defined by the Persistent Volume Claim specified during workbench creation. Limit ranges and memory limits control compute constraints, so expect PVC configuration when dealing with persistent workspace storage on the exam.

Correct answer: A. Navigate to the Pipelines section, select the Pipeline Run, click on the specific graph node, and view the Logs tab

Pipeline run logs are viewed by selecting the specific node in the pipeline run graph within the OpenShift AI dashboard. Using CLI tools or external terminals is incorrect because the dashboard visualizes execution logs directly per step.

Correct answer: B. Open the Terminal tool option from within the JupyterLab UI layout

You can access the terminal directly from the JupyterLab interface by opening a new terminal tab. You do not need to use external SSH clients or local terminal sessions for basic interactions within a workbench.

Correct answer: B. Configuring ResourceQuotas and LimitRanges on individual Data Science Project namespaces

Configuring LimitRanges and ResourceQuotas on Data Science Projects restricts resource consumption and prevents noisy neighbor issues. Adjusting permissions or storage buckets does not actively govern compute resources like CPU and memory.

Correct answer: C. Kubeflow and Open Data Hub platforms

Open Data Hub is the open-source community project that serves as the upstream foundation for Red Hat OpenShift AI. While related to Kubernetes, the specific upstream alignment is with Open Data Hub, not the core engine itself.

Correct answer: A. S3-Compatible Object Store storage

An S3-compatible object storage bucket is strictly required to store global pipeline artifacts and pass data between pipeline steps. File systems or block storage lack the required compatibility for the pipeline engine's storage backend.

Correct answer: D. Any S3-compatible object storage system like MinIO or AWS S3

OpenShift AI pipelines use MinIO or other S3-compatible systems for their object storage backend. This ensures flexibility across cloud or on-premises environments, whereas local paths and FTP lack the required API integration.

Correct answer: D. NodeSelector combined with AcceleratorProfiles

Accelerator profiles define exactly how hardware like GPUs are requested and consumed by a workbench pod. Compute resource requests alone only allocate generic resources, while the profile specifically handles the GPU integration.

Correct answer: B. It provides simultaneous multi-framework model execution across CPUs and GPUs

Triton Inference Server allows simultaneous deployment of models from multiple frameworks like TensorFlow and PyTorch. Other runtimes are often single-framework, whereas Triton standardizes multi-framework execution across hardware.

Correct answer: A. By accessing the target Prometheus/Grafana dashboards provided via OpenShift cluster monitoring integration

OpenShift AI integrates directly with the built-in cluster monitoring stack using Prometheus and Grafana. You can access these dashboards through the OpenShift web console to view operational metrics and component health.

Correct answer: D. DSCInitialization

The DSCInitialization custom resource manages global deployment properties for OpenShift AI. While the DataScienceCluster resource toggles components, DSCInitialization applies foundational settings required before those components can run.

Correct answer: B. supportedModelFormats

The supportedModelFormats field associates your runtime with specific model types so KServe or ModelMesh knows when to use it. Without this field, the platform cannot match incoming model deployment requests to your custom runtime.

Correct answer: D. OdhDashboardConfig Custom Resource

The OdhDashboardConfig custom resource manages dashboard settings and feature toggles. While the DataScienceCluster resource manages operator components, the OdhDashboardConfig is the specific target for customizing the UI and dashboard capabilities.

Correct answer: A. To allow the pod to schedule onto tainted nodes like dedicated GPU nodes

Configuring tolerations allows the pod to schedule onto tainted nodes like dedicated GPU nodes. Tolerations explicitly permit workloads to bypass node restrictions so they can access specialized hardware.

Correct answer: B. Automating data preparation, model training, and validation tasks

A Data Science Pipeline automates data preparation, model training, and validation tasks within an enterprise MLOps lifecycle. It replaces manual notebook execution to ensure reproducible and scalable machine learning workflows.

Correct answer: D. Red Hat OpenShift AI Operator

The Red Hat OpenShift AI Operator is the required package for installing and managing the OpenShift AI components on a cluster. Other operators like Pipelines or GitOps are dependencies or separate tools, not the primary installation method.

Correct answer: D. Open Data Hub (ODH)

Red Hat OpenShift AI is built upon the Open Data Hub upstream community project. While Kubeflow is historically related and partially integrated, Open Data Hub is the specific upstream foundation tested on the exam.

Correct answer: A. Integrated MLflow Tracking Server instances

The integrated MLflow Tracking Server is the supported tool for logging machine learning parameters, metrics, and artifacts. Prometheus and Jaeger are used for infrastructure observability, not tracking model training data.

Correct answer: A. An object storage Data Connection and a Pipeline Server

You must provision an object storage Data Connection and a Pipeline Server to enable Data Science Pipelines. The Pipeline Server uses this data connection to authenticate to S3 storage for saving pipeline artifacts and outputs.

Correct answer: D. HTTP/REST or gRPC communication protocols

HTTP/REST or gRPC communication protocols are typically used to send inference requests to a deployed model via OpenShift AI model serving. Standard network protocols like FTP or SMTP are completely irrelevant to runtime model inferencing.

Correct answer: D. Pipeline artifact storage paths via object storage configuration

Sequential pipeline steps share data using object storage paths configured for pipeline artifacts. Exam tip: know that cross-step state relies on S3-compatible storage and Data Connections rather than ephemeral disk or memory.

Correct answer: C. containerPort within runtime spec

A custom serving runtime defines its listening endpoint using the containerPort field in the runtime specification. Exam cue: KServe uses this declared port to route inference traffic, bypassing standard Service Mesh configurations.

Correct answer: A. KServe isolates models per pod while ModelMesh packs models per pod

KServe deploys single models per pod for isolation, while ModelMesh packs multiple models into shared pods for efficiency. Understanding this resource allocation distinction is critical for choosing the right serving runtime.

Correct answer: A. Multi-Model Serving Runtime configurations

Multi-Model Serving Runtime configurations allow several models to share a single pod's resources. Single-model serving isolates them, so select the multi-model runtime when optimizing for cluster resource efficiency.

Correct answer: D. Check the Pipeline Runs section in the dashboard

Execution history and artifacts for workflows are tracked in the Pipeline Runs section of the dashboard. Git history only tracks code changes, whereas pipeline runs log the actual inputs and outputs.

Correct answer: C. Serverless Inference via Knative

Serverless inference via Knative scales your deployed models down to zero pods to conserve cluster resources. The OpenShift Serverless operator enables this behavior, but Knative provides the actual serverless inference functionality being described.

Correct answer: D. The workbench pod is deleted but the persistent storage remains intact

Stopping a workbench deletes the running pod but preserves the attached persistent volume claim. This ensures your project files and data remain safely stored, allowing you to restart the workbench later without losing any progress.

Correct answer: D. KServe Single-Model Serving

KServe Single-Model Serving provides dedicated resources and advanced autoscaling for large language models. ModelMesh is designed for high-density, multi-model serving to share resources, making it unsuitable for massive LLMs.

Correct answer: D. ReadWriteMany (RWX) profile

The ReadWriteMany access mode allows a persistent volume to be mounted as read-write by multiple nodes simultaneously. ReadWriteOnce restricts access to a single node, preventing concurrent mounting across separate workbench instances.

Correct answer: B. OpenShift Serverless and OpenShift Service Mesh

KServe uses Knative Serving on OpenShift Serverless for autoscaling and routing, often paired with Service Mesh. GitOps and Pipelines are deployment tools, while Serverless handles the scaling logic required for endpoints.

Correct answer: B. The model name, the storage data connection, and the path to the model file within the bucket

Deploying a model requires its name, a configured data connection, and the storage path. Dockerfiles are unnecessary because standard serving runtimes load models directly from the specified bucket via your data connection.

Correct answer: D. The Pod is deleted, but the persistent volume claim (PVC) remains intact to preserve data

Stopping a workbench deletes the running pod but retains the PVC. This preserves your files and environment configuration for the next session. Beware of options claiming storage destruction, which only happens upon project deletion.

Correct answer: D. The artifact is uploaded to object storage and downloaded by the next step

Data science pipeline steps are ephemeral and share artifacts through object storage rather than direct memory access. The output of one step is uploaded to S3 and then downloaded by the subsequent step in the workflow.

Correct answer: C. AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY

OpenShift AI uses the S3 API standard for object storage connections, which maps credentials to the AWS environment variables. Workbenches read these keys to authenticate with the designated enterprise storage system.

Correct answer: A. KServe Serving Stack Platform

KServe is the designated serving stack for deploying large foundation models that require advanced features like scale-to-zero functionality. Conversely, ModelMesh is intended for managing multiple smaller models concurrently within a shared infrastructure.

Correct answer: A. OpenShift Serverless (Knative)

KServe relies on OpenShift Serverless, powered by Knative, to provide scale-to-zero capabilities. Service Mesh handles internal routing, but Knative directly manages the lifecycle and traffic-driven scaling decisions for serverless inference.

Correct answer: A. Notebook Controller deployment manager service instance

The Notebook Controller specifically handles the allocation and lifecycle of Jupyter notebook pods within OpenShift AI. While ModelMesh handles model serving, it does not manage the creation of user workbenches, making it the primary distractor.

Correct answer: C. Task is forcefully terminated by controller

When a pipeline step exceeds its configured timeout duration, the execution engine forcefully terminates the task. Pipeline steps do not automatically retry upon timeout, they simply fail and halt the workflow progression.

Correct answer: A. OdhDashboardConfig user group properties

Administrators use the OdhDashboardConfig custom resource to define specific user groups allowed to access Jupyter workbenches. Standard OpenShift OAuth manages general cluster authentication but does not restrict dashboard visibility specifically.

Correct answer: C. Red Hat OpenShift Service Mesh based on Istio

OpenShift AI relies on Red Hat OpenShift Service Mesh, which is based on Istio, to secure and manage traffic for model endpoints. Other meshes like Linkerd or Consul are not integrated into the OpenShift AI architecture natively.

Correct answer: B. view project permission role

The view role allows a user to see project assets like workbenches and pipelines without granting modification permissions. The edit role allows changes, while cluster-admin provides overarching administrative access beyond a single project.

Correct answer: D. OpenShift Ingress or Service Mesh Ingress Gateway

External traffic to a serverless InferenceService is routed securely through the OpenShift Ingress or Service Mesh Ingress Gateway. Internal ClusterIP addresses are not accessible externally, making them unsuitable for this task.

Correct answer: D. It manages cluster orchestration, scheduling, and distribution of tasks to Worker Nodes

In a Ray cluster, the Head Node acts as the central brain, managing cluster orchestration, scheduling, and the distribution of tasks to Worker Nodes. It does not execute the heavy compute training loops independently.

Correct answer: B. Accelerator Profiles custom resource configuration definition

Accelerator Profiles are used to define available GPUs and their specific allocation or tolerations for data science projects. Resource Quotas are a generic OpenShift feature but do not map specific hardware accelerators to user teams.

Correct answer: B. Data Science Project resource domain

A Data Science Project acts as a specialized Kubernetes namespace that logically groups resources like workbenches, pipelines, and model servers. ModelMesh is strictly a runtime for model serving, not a grouping mechanism for all project resources.

Correct answer: C. OpenShift Service Mesh (Red Hat OpenShift Service Mesh / Istio)

KServe requires OpenShift Service Mesh, built on Istio, to properly handle inference traffic routing and mutual TLS. Standard OpenShift routers or HAProxy do not provide the advanced mesh capabilities required by the KServe architecture.

Correct answer: B. Data Connection definition schema

A Data Connection securely stores the credentials and endpoint URL needed to access model files from an S3-compatible object storage bucket. Pipeline definitions lack the specific object storage secrets needed to pull runtime model artifacts.

Correct answer: A. Token-based authorization via ServiceAccount tokens

KServe inference services rely on Kubernetes ServiceAccount tokens to verify and authenticate client requests to prediction endpoints. Basic authentication or standard NetworkPolicies alone do not provide the expected token-based authorization mechanism.

Correct answer: B. Kubeflow Pipelines SDK library

Data scientists use the Kubeflow Pipelines SDK in Python to programmatically compile components and define pipelines. Jenkinsfiles and YAML manifests are infrastructure tools, whereas the SDK specifically constructs the directed acyclic graph.

Correct answer: D. annotations specifying the deployment mode (e.g., serving.kserve.io/deploymentMode)

Annotations such as serving.kserve.io/deploymentMode dictate whether the model is deployed via KServe or ModelMesh. Standard configuration parameters like image names or volume sizes do not control the underlying serving runtime selection.

Correct answer: D. ServingRuntime CR

The ServingRuntime Custom Resource specifies the container image, environment variables, and serving behavior. This allows administrators to define distinct runtimes for different frameworks instead of modifying the Workbench CR.

Correct answer: D. View pod logs in the OpenShift console or via oc logs

Administrators monitor failing model deployments by viewing pod logs via the OpenShift console or oc logs. Serving runtimes run in pods, so checking container logs is the fastest way to find runtime errors.

Correct answer: D. Elyra Pipeline Editor

The Elyra Pipeline Editor is integrated into OpenShift AI Jupyter images to allow users to visually build pipelines as a graph. Remember that Elyra is the specific tool for visual pipeline creation.

Correct answer: C. Enforcing native Kubernetes Role-Based Access Control policies

OpenShift AI leverages native Kubernetes Role-Based Access Control policies to restrict users to their authorized data science projects. Know that RBAC handles resource isolation.

Correct answer: A. OdhApplication Custom Resources

Third-party application tiles are registered on the OpenShift AI landing page using OdhApplication Custom Resources. You should know how to apply these YAML manifests to customize the dashboard.

Correct answer: D. Review the status of the operator deployment in the OpenShift console

The correct answer is correct because checking the operator deployment status in the OpenShift web console is the standard way to verify its health. The Kubeflow SDK or workbench pods rely on the operator already functioning properly.

Correct answer: C. OAuth Proxy sidecar container pod

The OAuth Proxy sidecar is injected into the notebook pod to intercept traffic and handle OpenShift authentication. While the Ingress Controller routes external traffic, it does not manage the granular user session authentication required for private notebooks.

Correct answer: B. Inside the configured S3-compatible object storage bucket

OpenShift AI Data Science Pipelines rely on the configured object storage bucket to store pipeline definitions, run artifacts, and step logs. This external storage ensures data is preserved independently of the ephemeral pipeline execution pods.

Correct answer: C. Elyra Pipeline Pipeline Editor

The Elyra Pipeline Editor is used to visually create pipelines within JupyterLab in OpenShift AI. While Tekton executes the underlying pipeline steps, Elyra is the visual editor used for pipeline creation.

Correct answer: B. The pipeline step pod is terminated or throttled by the OpenShift quota controller

OpenShift will throttle or terminate pods that exceed the configured resource limits or quotas for the namespace. The cluster will not automatically restart, and the pipeline definition is retained regardless of execution failures.

Correct answer: B. KubeRay Operator / Ray Cluster integrations

OpenShift AI integrates with the KubeRay operator to manage and coordinate distributed machine learning workloads like Ray clusters. Traditional Hadoop or Ansible are not used for native distributed training orchestration in this context.

Correct answer: D. Blue-Green / Canary routing via Service Mesh

Blue-green or canary deployments route traffic gradually to prevent downtime when updating inference models. Deleting old services beforehand causes outages, whereas traffic splitting ensures continuous availability during the model rollout transition.

Correct answer: C. Modify the component status to Managed or Removed in the Custom Resource

Administrators manage add-on components by setting their status to Removed in the DSC Initialization custom resource. Manually deleting pods is temporary because the operator will just recreate them to match the desired state.

Correct answer: D. KServe v2 Data Plane

OpenShift AI uses the KServe v2 Data Plane as its standard serving infrastructure. Although runtimes like Triton can execute models, KServe provides the underlying architectural framework for managing deployments.

Correct answer: C. Interactive exploratory data science and model training

Workbench containers are purpose-built for interactive exploratory data science and model training. They are explicitly not designed for high-throughput production web applications; model serving runtimes handle that production traffic.

Correct answer: C. The corresponding OpenShift namespace and all its assets are removed

Deleting a Data Science Project removes the corresponding OpenShift namespace and all its underlying assets. This cascading deletion ensures all associated workbenches, pipelines, and deployed models are cleaned up properly.

Correct answer: C. Ray Operator or Code Engine

OpenShift AI uses the Ray Operator or Code Engine to orchestrate distributed training clusters dynamically. For the exam, focus on Ray as the primary solution for scaling multi-node training workloads.

Correct answer: A. cluster-admin role permissions

Configuring cluster-wide accelerator profiles requires cluster-admin role permissions. This elevated access is necessary because accelerator profiles dictate hardware resource allocation across the entire OpenShift AI environment.

Correct answer: D. Push code changes to a shared Git repository and pull updates

To collaborate using Git, a data scientist pushes code changes to a shared repository so team members can pull updates. Exporting images or sharing live URLs bypasses version control and breaks standard collaboration workflows.

Correct answer: D. DSCInitialization Custom Resource object

The DSCInitialization Custom Resource object handles lower-level cluster setup items like web terminal configurations. This resource ensures necessary underlying components and namespace configurations are deployed properly.

Correct answer: B. KServe Serving Framework Platform Architecture

The KServe framework provides standard APIs for managing machine learning models as serverless containers. It handles inference scaling natively, whereas tools like Jenkins or Kafka focus on continuous delivery or streaming.

Correct answer: C. ModelMesh-serving-coordinator container within the serving architecture

The ModelMesh-serving-coordinator container is responsible for managing model placement and routing across the serving pods. In the exam, remember that standard KServe or ModelMesh deployments rely on this internal coordinator rather than external cluster managers.

Correct answer: A. AWS S3 and any fully S3-compatible object storage (e.g., MinIO, Red Hat Ceph/ODF)

AWS S3 and other S3-compatible object storage solutions are the standard for OpenShift AI data connections. Block storage and local file shares lack the necessary S3 API compatibility required by the platform's underlying SDKs.

Correct answer: C. An object storage-backed Pipeline Server

An object storage-backed Pipeline Server must be provisioned in a project before creating or executing data science pipelines. This server manages the Argo workflows and relies on object storage to persist pipeline artifacts and outputs.

Correct answer: B. Red Hat OpenShift Service Mesh Operator

Serverless model serving in OpenShift AI explicitly requires the Red Hat OpenShift Service Mesh Operator to handle traffic routing. Knative Serving builds upon this service mesh layer, making it a mandatory prerequisite for serverless deployments.

Correct answer: C. The container is terminated by the OOM killer mechanism

The OOM killer terminates containers when they exceed their defined memory limits to protect the node. Workbenches are standard pods, so standard OpenShift memory enforcement applies, rather than automatic volume expansion or live pod migration.

Correct answer: C. Edit the Data Connection resource within the project dashboard

Existing Data Connections are easily updated by editing the resource directly within the OpenShift AI project dashboard. Reinstalling the operator or recreating the entire Data Science Project is an unnecessary distractor that causes complete data loss.

Correct answer: B. S3 API protocol via HTTPS secure transmission channel

The Pipeline Server communicates with the S3 bucket artifact repository using the S3 API protocol over HTTPS. This ensures secure object storage access, whereas traditional file sharing protocols like NFS or SSH are not supported.

Correct answer: A. AcceleratorProfile

AcceleratorProfile is the custom resource used to define identifiers, tolerations, and resource limits for accelerators like GPUs. The DataScienceCluster custom resource manages the overall OpenShift AI components and not the specific hardware tolerances.

Correct answer: A. IAM Roles for Service Accounts (IRSA)

IAM Roles for Service Accounts allow cloud resources to be accessed securely using short-lived tokens linked to a service account. The other options do not provide native, passwordless AWS authentication mechanisms for OpenShift workloads.

Correct answer: B. ONNX or IR Format structures

The OpenVINO Model Server primarily processes models serialized in the Open Neural Network Exchange or Intermediate Representation formats. While other frameworks can be converted, ONNX and IR are the direct native inputs for this specific runtime.

Correct answer: D. /opt/app-root/src target directory space

Standard OpenShift AI Jupyter workbenches mount persistent storage to the /opt/app-root/src directory, keeping user files intact across restarts. Other directories like /root or /var/log are used for system processes and lack persistence guarantees.

Correct answer: A. DataScienceCluster

The DataScienceCluster custom resource defines the active components and global configuration of the OpenShift AI installation. DSCInitialization handles preliminary setup, but DataScienceCluster is the primary resource for component management.

Correct answer: C. Project Environment Variable Secrets configuration

OpenShift Secrets are designed to store sensitive information like database passwords securely and can be mapped as environment variables. Storing passwords in ConfigMaps, plain text, or code exposes them to unauthorized users.

Correct answer: B. Create a Custom Notebook Image setting in the OpenShift AI dashboard administrative view

Administrators can add proprietary images directly via the Custom Notebook Image settings in the OpenShift AI dashboard. Modifying low-level operator configmaps or cluster roles bypasses the supported configuration method.

Correct answer: B. Elyra runtime processor integrated with the OpenShift AI pipeline backend

The Elyra runtime extension converts visual notebook pipelines into runnable Tekton resources within OpenShift AI. Avoid distractors like Knative or Ansible, which handle serving or automation rather than pipeline compilation.

Correct answer: B. S3 Object Storage via Data Connections

Elyra pipelines persist intermediate artifacts using S3 object storage defined through OpenShift AI Data Connections. For the exam, remember that ConfigMaps or emptyDir volumes do not provide the required durable shared storage.

Correct answer: B. Scaling inference pods based on request volume including to zero

OpenShift Serverless scales deployed inference endpoints dynamically, including scaling to zero when idle. This is essential for managing compute costs, unlike compilation or storage provisioning which serve different roles.

Correct answer: A. By defining default resource requests and limits in the ServingRuntime or via Namespace quotas

OpenShift AI relies on standard Kubernetes resource limits and project namespaces to bound inference service usage. Distractors like external SaaS routing or disk swapping are not valid cluster management strategies for resource isolation.

Correct answer: C. Object Store S3 API endpoint standard

Data Connections map directly to S3-compatible object storage APIs for reading and writing artifacts. Practical cue: use boto3 with your Data Connection credentials rather than attempting file or REST model serving protocols.

Correct answer: C. LimitRange and ResourceQuota objects

Administrators use standard Kubernetes LimitRange and ResourceQuota objects to constrain compute resources for OpenShift AI workloads. Exam cue: ClusterResourceQuotas apply across multiple namespaces, while standard quotas bound individual projects.

Correct answer: D. v2 data plane specification (gRPC/REST)

ModelMesh uses the v2 data plane specification for standardizing inference requests via gRPC or REST. Knowing this protocol is vital when configuring custom runtimes and troubleshooting prediction payloads on the exam.

Correct answer: A. Adjust the notebook controller ConfigMap or Dashboard configuration

Workbench container sizes are adjusted by modifying the notebook controller ConfigMap or the dashboard configuration. Reinstalling the operator is unnecessary, as resource limits are managed through cluster-level configurations.

Correct answer: A. rhoai-users or standard self-provisioner role

Users need standard project creation roles, such as self-provisioner, to create data science projects. Cluster-admin is not required for basic project setup, which is a common trap in access control scenarios.

Correct answer: D. KFP Tekton compiler component

The KFP Tekton compiler translates visual Elyra pipelines into runnable Tekton resources. Knowing this component is crucial for troubleshooting pipeline submission failures directly from the workbench interface.

Correct answer: C. Kubeflow Notebook Controller

The Kubeflow Notebook Controller manages the lifecycle of workbenches in OpenShift AI. For the exam, remember that while ModelMesh handles serving, the Notebook Controller is specifically responsible for spinning up and managing your interactive Jupyter environments.

Correct answer: C. Red Hat Ecosystem Catalog and Quay.io

Red Hat hosts the pre-built notebook container images in the Red Hat Ecosystem Catalog and Quay.io. You can verify this during the exam by checking the image references used in the workbench deployment configurations.

Correct answer: D. oc command line interface utility

The oc command-line utility allows administrators to manage OpenShift AI resources directly. While the dashboard handles high-level tasks, you must use oc to interact with the underlying Kubernetes objects like secrets and routes.

Correct answer: C. Data Connection Config

A Data Connection acts as a Kubernetes secret storing your S3 endpoint and credentials securely. This allows notebooks and pipelines within the project to easily access the object storage without requiring manual environment variable configuration.

Correct answer: B. Self-provisioner or dedicated project creator rights in the OpenShift cluster

Creating a Data Science Project maps to an OpenShift namespace, requiring self-provisioner rights. Cluster-admin is incorrect because standard users can create projects if granted the self-provisioner role, which is the default in many clusters.

Correct answer: C. Namespaces via Data Science Projects

Data Science Projects map directly to OpenShift namespaces, providing isolation. NetworkPolicies handle traffic flow, not user workspace separation. Remember that dashboards group workbenches and resources by these projects.

Correct answer: A. PersistentVolumeClaim

Workbenches use PersistentVolumeClaims to attach storage for saving notebooks. Secrets and ConfigMaps handle credentials or configuration, not persistent data storage. Expect PVCs to remain after workbench deletion.

Correct answer: C. A Persistent Volume with ReadWriteMany (RWX) access mode configuration

ReadWriteMany access mode allows multiple pods to mount the same volume simultaneously. ReadWriteOnce restricts access to one node, preventing concurrent mounting by different workbenches. Expect shared storage setups to require RWX.

Correct answer: A. Tekton / Kubeflow Pipelines

OpenShift AI pipelines are built on Tekton and Kubeflow Pipelines. Jenkins and Argo are not the native engines for this feature. When answering pipeline questions, immediately associate the dashboard with Tekton custom resources.

Correct answer: D. Node Feature Discovery (NFD) Operator

The Node Feature Discovery Operator identifies and labels specialized hardware like GPUs in the cluster. While NVIDIA Lifecycle Manager handles driver provisioning, NFD is the core OpenShift component used for initial hardware detection and feature labeling.

Correct answer: B. Future steps needing artifact storage fail instantly

Pipeline steps rely on object storage for artifact passing, so removing the S3 connection breaks this dependency immediately. Subsequent tasks that require those inputs will fail when they attempt to download the missing artifacts.

Correct answer: B. dashboard.opendatahub.io/NotebookImage

Custom notebook images are defined using the NotebookImage custom resource within the dashboard.opendatahub.io API group. This allows administrators to programmatically provide custom environments for data scientists.

Correct answer: A. Kubernetes Secret object

OpenShift AI pipelines natively integrate with Kubernetes to store sensitive repository credentials in standard Secret objects. These secrets are then securely mounted or referenced by the pipeline tasks.

Correct answer: A. It specifies the target namespace where OpenShift AI core applications and components are installed

The DSCInitialization configuration sets up the required namespaces for the core OpenShift AI operators and their components. The applicationsNamespace field specifically dictates where the shared applications are deployed.

Correct answer: A. Knative Serving platform autoscaling component

Knative Serving provides the underlying autoscaling capabilities that allow KServe to scale model serving instances down to zero. Remember that ModelMesh is the alternative stack used for multi-model serving, whereas KServe handles large, single models via serverless features.

Correct answer: C. MariaDB or MySQL database instances provisioned within the infrastructure

The Data Science Pipelines application uses a MariaDB or MySQL database instance to store pipeline tracking metadata. While PVCs store the artifacts, the structured metadata for experiment tracking relies on this relational database backend.

Correct answer: C. Selecting an accelerator profile during workbench configuration creation

You must select an existing accelerator profile during the workbench creation process to allocate GPU resources to the notebook container. Simply installing Python libraries will fail if the underlying hardware profile is not requested.

Correct answer: D. ServingRuntime CRD instance

The ServingRuntime custom resource allows administrators to define and provision global runtime configurations for distinct model serving engines. An InferenceService acts as the actual instance pointing to a model, not the runtime environment itself.

Correct answer: C. Prometheus metric exposition format standards

KServe uses the Prometheus metrics exposition format to expose real-time model prediction statistics. While log streams contain generic runtime data, structured Prometheus metrics are required for proper scaling and monitoring integrations.

Correct answer: B. oc get pods -n openshift-operators

Checking the openshift-operators namespace retrieves the status of installed operators. Podman and systemctl do not inspect OpenShift cluster workloads, making oc get pods the correct administrative troubleshooting command.

Correct answer: B. Security Context Constraints (SCC) profiles

Security Context Constraints restrict pod permissions to prevent containers from hijacking host kernels. NetworkPolicies are also correct for network isolation, but SCCs strictly govern the container runtime permissions.

Correct answer: B. Red Hat OpenShift Local Storage Operator / OpenShift Data Foundation (ODF)

OpenShift AI pipelines require underlying persistent storage to function, typically provisioned using OpenShift Data Foundation or the Local Storage Operator. GitOps and Advanced Cluster Security do not provide the physical storage volumes required for pipeline artifacts.