Cloudera CDP-3002 flashcards
Cloudera CDP-3002 · 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.
- CDP deployment models
- CDP runs as Public Cloud, where Cloudera hosts the control plane and environments live in a customer cloud account, or as Private Cloud Base, where Cloudera Manager manages clusters on customer infrastructure.
- Data Hub cluster template
- A reusable definition of services, instance types, and node counts used to provision Data Hub clusters. Service composition is fixed at creation, so changing services requires a new template and cluster.
- Parcel activation order
- Download, distribute, activate, then restart services. A newly activated parcel does not take effect until the dependent service roles are restarted, and the old parcel can be deactivated afterwards.
- Gateway role
- A client configuration role placed on an edge node. It supplies client configs such as hive-site.xml and core-site.xml without running any service daemon, so users can submit jobs remotely.
- YARN queue capacity
- In the capacity scheduler, each queue has a guaranteed capacity and a maximum capacity. A queue can borrow unused capacity from others up to its maximum, and ACLs control who can submit.
- Cloudera Manager API port
- The Cloudera Manager REST API listens on port 7183 by default, using paths of the form /api/<version>/clusters/<name>/services for automation and monitoring.
- Base versus Data Hub upgrade
- Cloudera Manager upgrades the base cluster first, then Data Hub clusters can be upgraded to the matching runtime. The base cluster must be at a version that supports the target runtime.
- Hive metastore role
- The Hive metastore is the shared catalog of databases, tables, partitions, and locations. Impala, Hive, and Spark read the same definitions, which is why DDL in one engine affects the others.
- Impala REFRESH versus INVALIDATE
- REFRESH reloads metadata for a known table after data or partition changes. INVALIDATE METADATA reloads all metadata and is used when tables were created or dropped outside Impala.
- Iceberg time travel
- Iceberg tables keep snapshots, so queries can use AS OF TIMESTAMP or VERSION AS OF to read historical data. Writes are ACID and hidden partitioning avoids partition columns in queries.
- COMPUTE STATS
- An Impala statement that gathers table and column statistics used by the planner. Without statistics, the planner may choose a poor join strategy such as a shuffle join instead of a broadcast join.
- Spark cluster deploy mode
- In cluster mode the driver runs inside the YARN ApplicationMaster, so the submitting client can disconnect after submission. In client mode the driver runs on the submitting host.
- Hive ACID table requirements
- Transactional Hive tables must be stored as ORC, bucketed, and marked transactional=true. They support INSERT, UPDATE, DELETE, and MERGE, but only compatible engines can read them.
- MSCK REPAIR TABLE
- A Hive command that scans HDFS for partition directories and adds any missing partitions to the metastore. It is used after files are loaded directly into HDFS.
- Oozie coordinator job
- A coordinator schedules workflow executions based on time frequency and data availability. It references a workflow application and defines the dataset and input events that trigger runs.
- Broadcast join
- An Impala join strategy where the smaller build side is sent to every node. It is efficient when one side is small enough to fit in memory on each executor.
- Ranger policy evaluation
- Ranger evaluates allow and deny policies using the configured order, either deny-override or allow-override. Resources with no matching policy are denied by default.
- Ranger versus Sentry
- In CDP Private Cloud Base, Ranger provides authorization for Hive and Impala through a shared Hive service definition, replacing Sentry used in older CDH and HDP releases.
- HDFS encryption zone
- A directory whose contents are encrypted with a key from the Key Management Server. Files cannot be moved into or out of a zone; they must be copied.
- Atlas tag-based policies
- Atlas classifies metadata with tags, and Ranger can reference those tags in policies. Tag synchronization between Atlas and Ranger must be enabled for the policies to apply.
- Kerberos prerequisite
- Kerberos authentication must be enabled before features such as HDFS transparent encryption and Ranger authorization can be configured on a secure CDP cluster.
- Auto-TLS in Cloudera Manager
- Cloudera Manager can generate an internal certificate authority and issue certificates for cluster hosts. Alternatively, certificates from an external CA can be imported.
- Hue SQL authorization
- Hue enforces Hive and Impala authorization through Hadoop SQL policies and HDFS access through Ranger policies. The logged-in user's group memberships determine the effective policy set.
- Ranger audit destination
- Ranger audit events can be written to HDFS, Solr, or a database. The audit store must be configured and healthy, otherwise policy decisions are logged but not persisted.
- Cloudera Manager health test
- A periodic check run per role. A red icon means a health test failed, while a yellow icon indicates a warning or an in-progress command that needs attention.
- Impala out-of-memory error
- An impalad memory limit exceeded during a query. Remedies include raising the memory limit and reducing query concurrency or the size of joins and sorts.
- YARN physical memory limit
- A container that exceeds its physical memory allocation is killed with a message about running beyond physical memory limits. Adjust container size or NodeManager memory.
- HDFS safe mode
- A NameNode state that blocks writes while it waits for block reports. Use hdfs dfsadmin -safemode get and leave after verifying the NameNode heap and block reports.
- Diagnostic bundle
- A Cloudera Manager artifact that collects logs, configurations, and metrics from cluster hosts. It is the standard package attached to a support case.
- GSS initiate failed
- A Kerberos error indicating missing or expired credentials. Check the keytab, ticket lifetime, and clock skew between hosts, which must stay within about five minutes.
- HDFS snapshot
- A point-in-time, read-only view of a directory used for recovery from accidental deletion or change. Snapshots are enabled per directory and stored in the NameNode.
- DistCp for disaster recovery
- A Hadoop tool that copies data between clusters using MapReduce or YARN. It is used for cross-cluster replication, while the metastore database is backed up separately.