Cloudera CDP-3002 key terms

The web-based control plane that installs, configures, monitors, and starts or stops services across a CDP cluster. It tracks host health, raises alerts, and stores configuration history for rollback.

A versioned binary distribution of Cloudera Runtime that Cloudera Manager distributes and activates on hosts. Parcels allow side-by-side versions and rollback without reinstalling operating-system packages.

The bundled set of open-source components (HDFS, YARN, Hive, Impala, Spark, and others) shipped as a parcel and managed together by Cloudera Manager on CDP hosts.

A configuration file inside a parcel that declares a service's roles, configuration parameters, and dependencies, letting Cloudera Manager generate and validate configuration for that service.

A named collection of role instances that share one configuration set in Cloudera Manager. Editing a role group changes every member at once, simplifying management of large clusters.

A client-side role that deploys service client configuration files to an edge host so tools and applications there can reach cluster services without running server daemons.

An HDFS deployment with active and standby NameNodes sharing an edit log through JournalNodes. A failover controller promotes the standby if the active node fails.

A lightweight HDFS daemon that stores shared edit-log entries so multiple NameNodes can stay consistent. Quorum write guarantees let the standby take over without data loss.

A design that runs several independent NameNodes, each owning a namespace and block pool, so HDFS capacity and namespace load scale beyond a single NameNode.

The cluster-wide scheduler that allocates resources to applications and launches ApplicationMasters. It enforces queues, capacity limits, and fairness policies defined in scheduler configuration.

The per-host YARN agent that reports available memory and CPU, launches containers on request, and enforces container limits. It also runs auxiliary services and local log aggregation.

A hierarchical YARN queue with configured capacity, maximum capacity, and user limits. Applications submit into a queue, and the scheduler shares resources among queues by policy.

The unit of resource allocation in YARN: a granted amount of memory and virtual cores on one NodeManager where a task process runs until it completes or is killed.

The service that stores table definitions, partitions, and locations for Hive, Impala, and Spark. Losing or misconfiguring it makes existing tables invisible to query engines.

A table whose data files live outside the managed warehouse directory. Dropping the table removes metadata only; the underlying files remain and can be re-registered later.

A table whose data files sit in a location controlled by the metastore. Dropping it deletes both metadata and stored data, so it suits disposable or fully governed datasets.

A table format storing data as Parquet files plus metadata and snapshot manifests. It supports ACID commits, schema evolution, time travel, and hidden partitioning on CDP engines.

A query optimization where the engine reads only partitions matching filter predicates, skipping unrelated directories or files. Good partition column choice sharply cuts scanned data.

A massively parallel SQL engine that runs queries directly on HDFS and object-store data without MapReduce. It relies on the metastore catalog and uses stateless coordinators and executors.

Running Spark applications with YARN as the cluster manager, where an ApplicationMaster requests executors. Executor count and memory are tuned through spark-submit properties.

A workflow scheduler for Hadoop jobs that defines actions and dependencies in XML. It coordinates Hive, Spark, and shell steps into ordered workflows and recurring coordinators.

A directed graph of dependent jobs with defined start and end nodes. In CDP tooling it captures execution order, retries, and parameters so pipelines run reproducibly.

A flow-based data movement tool with processors, connections, and back-pressure. It ingests, routes, and transforms streaming or batch data while tracking provenance for each record.

A distributed publish-subscribe log organized into partitioned topics with replicated brokers. Producers append records and consumer groups read them at their own offsets.

The centralized authorization service for CDP. It stores policies per resource such as HDFS paths, Hive tables, or Kafka topics and enforces them through plugins in each component.

A rule granting or denying access to a resource for users, groups, or roles, optionally with row filters, column masks, and time conditions. Policies are evaluated in priority order.

A Ranger or Hive feature that rewrites a query so a user sees only rows matching a predicate, such as their own region, without changing the underlying table data.

A policy that replaces sensitive column values with redacted or hashed output for selected users at query time, leaving stored data unchanged for authorized readers.

The metadata and governance catalog in CDP. It captures lineage, classifications, and business metadata for datasets and integrates with Ranger for tag-based access control.

An access rule attached to a classification such as PII rather than to a physical path or table. Any asset Atlas tags with that classification inherits the restriction automatically.

A unique identity in the Kerberos realm, usually service/host@REALM for daemons or user@REALM for people. Services authenticate to each other by proving possession of keytabs.

A file holding encrypted long-term keys for one or more Kerberos principals. Daemons and scheduled jobs use it to authenticate without an interactive password.

A short-lived credential issued after Kerberos authentication so a distributed job's tasks can access services without forwarding the user's ticket or keytab.

A signed public-key document that lets CDP services encrypt traffic and verify each other's identity. Expired or mismatched certificates break agent heartbeats and web UI access.

A notification triggered when a health check crosses a threshold, such as low disk or a down role. Alerts carry severity and can forward to email or external monitoring.

A named test Cloudera Manager runs against a service or role, returning GOOD, CONCERNING, or BAD. Its thresholds and enablement are configurable per role group.

A collected archive of logs, configuration, and metrics from cluster hosts, generated by Cloudera Manager to support troubleshooting without manual log gathering.

A Cloudera Manager or Replication Manager rule that copies HDFS data, Hive metadata, or snapshots from a source to a target cluster on a schedule for disaster recovery.

A read-only point-in-time view of a directory that consumes little extra space by sharing unchanged blocks. It enables rollback and consistent copies for replication.