CDP (Big Data Hadoop Administrator) 5 — Questions and Answers
Question 1: A cluster shows that one DataNode's disk usage is at 95% while others are at 40%. Which parameter controls the threshold at which a DataNode is considered full and stops accepting new blocks?
- dfs.datanode.du.reserved (Correct answer)
- dfs.datanode.disk.check.timeout
- dfs.namenode.disk.check.min.gap
- dfs.datanode.failed.volumes.tolerated
Correct answer: dfs.datanode.du.reserved
dfs.datanode.du.reserved reserves a fixed number of bytes on each DataNode volume so the node stops accepting blocks before the disk is completely full.
Question 2: Which Hadoop service provides a REST gateway for interacting with HDFS from non-Java clients?
- Hue
- WebHDFS / HttpFS (Correct answer)
- Ambari
- Knox
Correct answer: WebHDFS / HttpFS
WebHDFS is built into the NameNode and exposes an HTTP REST API; HttpFS is a standalone proxy offering the same API with better firewall-friendliness.
Question 3: In HDFS, what is a 'rack-aware' block placement policy designed to achieve?
- Maximizing write throughput by placing all replicas on the same DataNode
- Balancing fault tolerance and bandwidth by spreading replicas across racks (Correct answer)
- Ensuring blocks are always placed on the DataNode closest to the client
- Reducing NameNode metadata size by grouping blocks per rack
Correct answer: Balancing fault tolerance and bandwidth by spreading replicas across racks
Rack-aware placement puts one replica locally, one on a different rack node, and one more on that second rack so a single rack failure doesn't lose data.
Question 4: Which command forces the NameNode to exit safe mode immediately without waiting for the threshold to be met?
- hdfs dfsadmin -safemode get
- hdfs dfsadmin -safemode leave (Correct answer)
- hdfs dfsadmin -safemode enter
- hdfs dfsadmin -safemode forceExit
Correct answer: hdfs dfsadmin -safemode leave
hdfs dfsadmin -safemode leave forces the NameNode out of safe mode regardless of whether the minimum replication threshold has been satisfied.
Question 5: What is the role of the HDFS Checkpoint Node (formerly Secondary NameNode)?
- Serving as a hot standby that takes over if the primary NameNode fails
- Periodically merging the fsimage and edit log to create a new checkpoint (Correct answer)
- Balancing read requests across multiple NameNodes
- Storing encrypted keys for HDFS Transparent Encryption
Correct answer: Periodically merging the fsimage and edit log to create a new checkpoint
The Checkpoint Node (Secondary NameNode) regularly combines the current fsimage and accumulated edits into a new fsimage, preventing the edit log from growing indefinitely.
Question 6: Which YARN property configures the maximum amount of memory that a NodeManager can allocate to all containers running on that node?
- yarn.nodemanager.vmem-pmem-ratio
- yarn.nodemanager.resource.memory-mb (Correct answer)
- yarn.scheduler.maximum-allocation-mb
- mapreduce.map.memory.mb
Correct answer: yarn.nodemanager.resource.memory-mb
yarn.nodemanager.resource.memory-mb sets the total physical memory the NodeManager can offer to YARN containers on that host.
Question 7: What is Hadoop's 'short-circuit reads' feature and which configuration enables it?
- Reading HDFS blocks directly from disk without going through the DataNode daemon, enabled via dfs.client.read.shortcircuit (Correct answer)
- Bypassing replication for local writes, enabled via dfs.replication.min=0
- Allowing NameNode to read blocks directly from client memory, enabled via dfs.namenode.shortcircuit
- Caching frequently accessed blocks in RDMA memory, enabled via dfs.datanode.rdma.enabled
Correct answer: Reading HDFS blocks directly from disk without going through the DataNode daemon, enabled via dfs.client.read.shortcircuit
Short-circuit reads let a co-located client read HDFS block files directly from local disk via a Unix domain socket, bypassing DataNode network overhead.
CDP (Big Data Hadoop Administrator) 5
A cluster shows that one DataNode's disk usage is at 95% while others are at 40%.
Which parameter controls the threshold at which a DataNode is considered full and stops accepting new blocks?