IBM Cloud Pak for Data v4.7 Architect key terms

The shared set of platform services that manage identity, cataloging, metering, and the web client, deployed once per cluster so that installed services can operate together under a single namespace.

A per-service installation managed by an operator, separate from the control plane, that can be scaled, upgraded, or removed independently while sharing platform-wide identity and catalog services.

The Kubernetes subsystem that installs, upgrades, and tracks operators and their subscriptions, providing the mechanism by which Cloud Pak for Data services are deployed and versioned on the cluster.

A registry-backed index of operator bundles that the cluster queries to discover available Cloud Pak for Data versions and channels during installation or upgrade.

A subscription setting that determines whether operator upgrades install automatically or wait for an administrator to approve them, controlling when a new service version reaches the cluster.

The supported sequence of versions a deployment must follow to reach a target release, often requiring intermediate releases or specific control plane upgrades before service operands.

The process of capturing platform state, including databases and persistent volumes, so a deployment can be recovered after failure or rolled back when an upgrade fails.

A request for storage bound to a physical volume, used by stateful Cloud Pak for Data components to retain data across pod restarts and rescheduling.

A cluster-defined profile describing how dynamic volumes are provisioned, including performance and access modes, which must satisfy the read-write requirements of platform services.

Scheduling rules that constrain pods to particular nodes based on labels, commonly used to place GPU or high-memory workloads on appropriately sized worker nodes.

A namespace-level limit on aggregate CPU, memory, or storage consumption that prevents one project from exhausting cluster capacity needed by platform services.

A pairing where a node repels pods unless the pod declares a matching toleration, used to reserve dedicated nodes for licensed or resource-intensive Cloud Pak for Data workloads.

The identity and access microservice layer of the platform that authenticates users, issues tokens, and enforces role-based permissions across all installed services.

A permission model where users are assigned platform or service roles that grant specific capabilities, such as viewing a catalog or deploying a model, rather than individual permissions.

A policy model that grants or denies access based on characteristics of the user, resource, and environment, evaluated at query time rather than through static role assignment.

Configuration that lets users authenticate with enterprise credentials through SAML or OIDC, mapping external groups to platform roles instead of maintaining separate local accounts.

A long-lived credential generated for a user or service that authorizes programmatic calls to platform APIs without embedding a password in scripts or notebooks.

A non-human identity used by applications and automated pipelines to obtain tokens and access platform resources under controlled, auditable permissions.

The component that intercepts a data request and applies governance decisions, blocking, masking, or allowing access before results are returned to the requester.

Governance policies that mask, filter rows, or otherwise transform data at access time so sensitive values are hidden from users who lack the required clearance.

A curated collection of business terms with definitions, owners, and relationships that gives shared meaning to data assets across the organization.

A recorded map of how data moves and transforms from source systems through jobs and models to reports, used for impact analysis and audit.

The process of harvesting technical metadata from connected sources into the catalog so assets become searchable, assignable to terms, and subject to governance rules.

A stored, reusable definition of how to reach a data source, including endpoint, credentials, and driver, referenced by integration, virtualization, and cataloging assets.

A technique that exposes remote tables as queryable assets without copying data, letting users join sources while governance policies are applied to results.

Querying multiple heterogeneous sources through a single virtual layer that pushes down compatible operations and combines results, avoiding full data replication.

Delegating filters, joins, and aggregations to the remote source so less data crosses the network, which strongly affects performance of virtualized queries.

The planner's choice of join order, pushdown, and execution strategy for a virtual or integrated query, often tunable through statistics and configuration hints.

Copying source data into a target store on a schedule or change stream, trading storage and latency for faster local access and reduced load on source systems.

Reading insert, update, and delete events from source logs to keep a target synchronized incrementally instead of reloading entire tables.

A pipeline step that extracts, transforms, and loads data, chained with other stages to build repeatable integration jobs with error handling and scheduling.

A collaborative workspace where users add data, notebooks, jobs, and models, with its own membership and access controls scoped to that work.

A governed collection of data and AI assets that users can search, preview, and request access to, with metadata and terms attached for discovery.

An environment paired with compute resources where models and functions are deployed and served, isolating runtime workloads from authoring projects.

A served model endpoint that answers synchronous requests with low latency, used by applications that need real-time predictions.

A scheduled or on-demand scoring job that processes a large dataset and writes predictions to a target, suited to throughput over latency.

The governance and cataloging service that provides asset discovery, glossaries, lineage, quality analysis, and policy enforcement across connected data sources.

The authoring service where users build notebooks, visual flows, and machine learning models, then promote them to deployment spaces for serving.

Collecting cluster and namespace diagnostics, logs, and resource descriptions into an archive for support analysis when a service fails to start or behave correctly.