FusionInsight and Big Data Operations
This domain tests day-to-day operation of a FusionInsight HD cluster through FusionInsight Manager: services, roles and instances, alarms, health checks, tenants, capacity, backup, and upgrades. Questions usually describe a symptom and ask for the correct operational response.
Core rules
- Manager is the unified console. A cluster contains services; each service contains roles distributed as instances across nodes.
- Multi-tenant management allocates compute queues, HDFS storage directories, and HBase namespaces to tenants. Resource changes affect scheduling, so plan before changing quotas.
- HA covers NameNode, ResourceManager, and other critical roles. Backups use snapshots and metadata copies; snapshots are not a substitute for off-cluster backup.
- Routine operations: review alarms, run health checks, monitor capacity and under-replicated blocks, rebalance DataNodes, expand or upgrade with vendor procedures, and audit Kerberos-secured access.
Worked example
Manager raises a DataNode disk-usage alarm at 85%. Reasoning: first check whether usage is cluster-wide or one node; run a rebalance if skewed; if the cluster is genuinely near capacity, add DataNodes or archive cold data through proper tools. Do not delete block files from the DataNode filesystem, because that corrupts replicas. After remediation, verify that the alarm clears and under-replicated blocks return to zero.
Traps
- Restarting the whole cluster is not a first response to a single-role alarm; restart the affected instance after checking logs.
- Clearing an alarm in Manager does not prove the root cause is fixed; confirm metrics and block health.
- Deleting HDFS files directly on disk bypasses HDFS metadata and damages the cluster.