Platform provisioning and cluster management
Domain 1 tests the CDP Public Cloud object model and the mechanics of creating and operating environments, a Data Lake, and Data Hub clusters. Questions usually state a requirement (high availability, isolation, cost control) and ask which construct, template, or CLI action satisfies it.
Environment, Data Lake, Data Hub
An environment binds a cloud account and region to a VPC, subnets, an SSH key, and storage locations, and it can hold exactly one Data Lake. The Data Lake is not just another cluster: it runs the SDX services (Ranger, Atlas, Hive Metastore, IDBroker, FreeIPA/Kerberos) and uses object storage as its filesystem, so every Data Hub in that environment inherits authentication, authorization, and the metastore. Data Hub clusters are disposable compute attached to that shared foundation.
Provisioning prerequisites are a cloud credential (AWS cross-account IAM role, Azure service principal, GCP service account), a network with enough private subnets, an SSH key pair, and separate data and logs buckets. Data Lake HA requires three subnets in three availability zones.
Templates, definitions, lifecycle
A cluster definition combines a versioned template (services, instance types, scaling rules) with the cloud provider. Templates can be cloned and customised, and bootstrap recipes are bound at creation time. Data Hubs support stop/start, manual resize, autoscaling, repair, and runtime upgrade. The Data Lake cannot be stopped, and its upgrades are orchestrated by Cloudera rather than self-service.
Worked example
Requirement: a production Data Lake that survives one AZ failure, plus a dev cluster reading the same schemas. Reasoning: create one environment with three private subnets in distinct AZs and provision the HA Data Lake; place the dev Data Hub in the same environment so it shares the metastore and Ranger policies, accepting that dev workloads can reach production tables unless Ranger policies separate them. Reject two Data Lakes in one environment, and reject a single-AZ network for an HA Data Lake.
Traps
Assuming HDFS exists; embedding long-lived cloud access keys on nodes instead of using IDBroker credential vending; expecting to stop the Data Lake for cost savings; trying to attach a recipe after creation; placing subnets without NAT or object-storage gateway endpoints so bootstrap hangs.