IBM Cloud Pak for Data v4.7 Architect flashcards
IBM Cloud Pak for Data v4.7 Architect · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Cloud Pak for Data control plane
- Zen provides the home page, IAM integration, and platform API. If Zen is unhealthy, the platform UI is unavailable even when individual data services are still running.
- Common core services role
- Common core services supply shared MongoDB, RabbitMQ, and search backing stores used by many services. Their failure affects multiple services at once, so monitor them as shared dependencies.
- Service versus service instance
- A service is the installed operator and images. An instance is a deployed copy with its own persistent volumes and resource requests, so capacity planning is per instance.
- Production cluster topology rule
- Production requires at least three master and three worker nodes spread across zones with anti-affinity. A single-zone cluster remains a single point of failure regardless of pod replicas.
- RWX versus RWO storage
- Shared platform services need ReadWriteMany file storage such as NFS or OCS. Block storage limited to ReadWriteOnce suits only single-pod workloads and breaks multi-replica services.
- GPU node isolation
- GPU nodes are needed only for NLP, speech, and foundation-model inferencing. Use taints and node selectors so ordinary workloads do not consume expensive GPU capacity.
- Entitlement key requirement
- The IBM entitlement key must be added to the global pull secret before installing components. Without it, component pods fail with ImagePullBackOff against the entitled registry.
- Operator and instance namespaces
- Operators are installed in cpd-operators and platform or service instances in cpd-instance. Keeping them separate simplifies RBAC, backup, and troubleshooting during upgrades.
- cpd-cli manage install order
- The order is prerequisites, platform (zen), services, then service instances. Commands are idempotent, so a failed step is retried rather than reinstalling the whole platform.
- Upgrade sequencing rule
- Upgrade the platform before services, never skip a required intermediate release, and confirm versions against the support matrix. Rollback means restoring the pre-upgrade backup.
- Pre-upgrade backup targets
- Back up MongoDB, Zen metadata, and catalog data to S3-compatible storage before upgrading. Validate the restore path, because rollback depends entirely on that backup.
- Air-gapped installation approach
- Mirror all images to a private registry with cpd-cli manage mirror-images, then set the image registry prefix. The entitlement key is still required at mirror time.
- Install health verification
- Check operand status with cpd-cli manage get-cr-status and runtime state with oc get pods. A component is not ready while its operand CR reports a non-Completed status.
- External identity provider integration
- Authentication is delegated to LDAP, SAML, or OIDC. The platform stores identity mappings and group-to-role assignments, not enterprise user passwords.
- Layered authorization model
- Access requires the intersection of platform role, service role, and resource role. Granting a platform role alone does not give access to catalog assets or project data.
- Data protection rules
- Watson Knowledge Catalog policies are enforced by rules. A data protection rule can deny access, mask column values, or filter rows at query time for governed assets.
- Platform role definitions
- Administrator manages users and services, User creates projects and instances, and Viewer has read-only UI access. These are distinct from per-service and per-resource roles.
- Certificate replacement procedure
- Replace the self-signed platform certificate with an enterprise CA-signed certificate using cpd-cli. Clients then trust a single CA chain across all services.
- Audit export to SIEM
- Platform and service audit records are exported to a SIEM through audit logging configuration. Catalog activity is recorded per asset for compliance review.
- Credential storage rule
- Store data source credentials in the platform credential store or an external vault. Never hardcode credentials in notebooks, job parameters, or connection strings.
- Watson Query nickname behavior
- A nickname is a virtual table over a remote source. Operations push down when the source supports them; cross-source joins require staging data locally.
- Virtualization versus DataStage
- Use virtualization for read-mostly federation without copying data. Use DataStage when transformation, cleansing, or a persisted target table is required.
- Data Replication use case
- Data Replication performs change data capture for near-real-time synchronization. Choose it when latency requirements are seconds to minutes, not batch ETL.
- Lineage capture scope
- Lineage is captured automatically for DataStage, Watson Query, and notebooks. External tools appear in lineage only after metadata import.
- Discovery and metadata import
- Discovery jobs profile assets, assign data classes from column names and values, and feed quality analysis. Results drive governance rules and catalog enrichment.
- DataStage partitioning and skew
- Partitioning method, buffering, and redistribution stages determine parallel job throughput. Skewed partitions are the most common cause of slow parallel jobs.
- Watson Query tuning levers
- Increase virtualizer resources, keep predicates pushdown-friendly, and pre-stage large cross-source joins. Local staging storage must be sized for those joins.
- DataStage parameter sets
- Parameter sets externalize connection strings and file paths so the same job runs across development, test, and production without editing job logic.
- ImagePullBackOff diagnosis
- ImagePullBackOff means the entitlement key or pull secret is wrong or expired. CrashLoopBackOff instead points to configuration, resource limits, or storage permissions.
- Pending PVC diagnosis
- A PVC stuck in Pending means no storage class satisfies the requested access mode. Inspect events and the storage class volumeBindingMode before changing the claim.
- Instance not ready causes
- A service instance that never becomes ready is often blocked by resource quota. Check quota and the operand CR status message before attempting a reinstall.
- Evidence collection before restart
- Gather oc logs, oc describe, oc get events, and cpd-cli collect-logs before restarting pods. Restarting destroys the failing state that support needs.