Big Data Architecture and Huawei FusionInsight Overview
What this domain covers
FusionInsight HD is Huawei's Hadoop-based big data platform; the exam checks that you can place each service in the data path and explain how Manager, security, and multi-tenancy fit around it. Core data services include HDFS for storage, YARN for resource scheduling, MapReduce/Tez/Spark for batch, Hive for SQL, HBase for random access, Flink for streaming, Kafka for buffering, and Flume/Loader/Sqoop for ingestion. ZooKeeper supplies coordination; Kerberos and Ranger supply authentication and authorization; FusionInsight Manager deploys, monitors, alarms, and manages tenants and services.
Rules to remember
- Manager is the management plane, not the data plane: client reads and writes go to HDFS, HBase, HiveServer2, Kafka brokers, or Flink REST endpoints, not through Manager.
- Master roles are usually active/standby with HA; data roles scale out horizontally.
- Batch relational ingestion fits Loader or Sqoop; continuous log ingestion fits Flume; event streams fit Kafka plus Flink.
- A service is not usable until its role instances are configured on nodes and started.
Worked example
A retailer needs order files from a relational database loaded into Hive every night, and clickstream events summarized in HBase within seconds. Reasoning: nightly relational extract is batch, so use Loader or Sqoop into HDFS/Hive; clickstream is continuous, so producers write to Kafka, Flink consumes and aggregates, then writes HBase. Hive serves the historical SQL layer; HBase serves point queries. Manager only monitors and configures these flows.
Traps that cost marks
Do not say Manager stores HDFS blocks. Do not use Flume for a one-off relational export. Do not assume adding a service automatically grants tenant access; Ranger policies and YARN queue assignments are separate.