I85PREP
⌘ K
Cloudera CDP-3002 study materials
Cloudera CDP-3002 study resources

Cloudera CDP-3002: Cloudera Data Platform Administration

Cloudera CDP-3002 · Work through your exam, one topic at a time.

All guide sections

Cloudera Data Platform Administration

The Cloudera Data Platform Administration domain tests how you install, configure, and scale a CDP cluster with Cloudera Manager, parcels, and service roles. Questions focus on the control plane, service dependencies, high availability, and safe change management rather than on pipeline code.

Control plane and parcels

  • Cloudera Manager Server stores configuration in its database; Agents on each host execute commands. A host must have a heartbeating Agent and pass inspection before roles can be assigned.
  • Parcels are the distribution mechanism: download, distribute, activate, then restart. Distribution copies the parcel to hosts; activation makes it the active version but does not restart services.
  • Configuration changes are versioned. Cloudera Manager marks services stale when a config or parcel change requires a restart. Use “Restart Stale Services” instead of a blind full-cluster restart.

Service dependencies and HA

  • HDFS underpins YARN, Hive, Impala, HBase, and Oozie. ZooKeeper is required for HDFS NameNode HA, YARN ResourceManager HA, and HBase.
  • HDFS HA uses two NameNodes, an odd number of JournalNodes (at least three), and ZKFC failover controllers. Clients use the nameservice ID, not a single hostname.
  • YARN HA uses two ResourceManagers coordinated through ZooKeeper. NodeManagers register with the active ResourceManager. Capacity Scheduler queues define memory and vCore limits.
  • To add a host: install the Agent, deploy client configuration, run the host inspector, then assign DataNode or NodeManager roles.

Worked example

A cluster has one NameNode and three JournalNodes, but clients still connect to hdfs://nn1:8020. You enable HA and add nn2.

Reasoning: create a nameservice ID, add the standby NameNode and ZKFC roles, bootstrap the standby from the active NameNode, start ZKFC, then fail over and verify. Update client configuration to the nameservice URI. Never format the active NameNode: formatting a NameNode that already holds namespace metadata can destroy the namespace.

Traps

  • Activating a parcel before it is distributed to every host.
  • Treating a rolling restart as universally safe during upgrades; role ordering matters.
  • Assuming a new DataNode automatically rebalances existing data. HDFS Balancer must be run.
  • Confusing replication factor with rack awareness: replication alone does not guarantee rack diversity without a topology script.