[ https://issues.apache.org/jira/browse/HDFS-9421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15007714#comment-15007714 ]
Mingliang Liu commented on HDFS-9421: ------------------------------------- Thanks for your reporting this issue, reviewing and committing the patch, [~xyao]! > NNThroughputBenchmark replication test NPE with -namenode option > ---------------------------------------------------------------- > > Key: HDFS-9421 > URL: https://issues.apache.org/jira/browse/HDFS-9421 > Project: Hadoop HDFS > Issue Type: Bug > Components: benchmarks > Affects Versions: 2.8.0 > Reporter: Xiaoyu Yao > Assignee: Mingliang Liu > Fix For: 2.8.0 > > Attachments: HDFS-9241.000.patch > > > Hit the following NPE when reviewing fix for HDFS-9387 with manual tests as > NNThroughputBenchmark currently does not have JUnit tests. > > {code} > HW11217:centos6.4 xyao$ hadoop > org.apache.hadoop.hdfs.server.namenode.NNThroughputBenchmark -op replication > -namenode hdfs://HW11217.local:9000 > 15/11/12 14:52:03 INFO namenode.NNThroughputBenchmark: Starting benchmark: > replication > 15/11/12 14:52:03 ERROR namenode.NNThroughputBenchmark: > java.lang.NullPointerException > at > org.apache.hadoop.hdfs.server.namenode.NNThroughputBenchmark$ReplicationStats.generateInputs(NNThroughputBenchmark.java:1312) > at > org.apache.hadoop.hdfs.server.namenode.NNThroughputBenchmark$OperationStatsBase.benchmark(NNThroughputBenchmark.java:280) > at > org.apache.hadoop.hdfs.server.namenode.NNThroughputBenchmark.run(NNThroughputBenchmark.java:1509) > at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:70) > at org.apache.hadoop.util.ToolRunner.run(ToolRunner.java:84) > at > org.apache.hadoop.hdfs.server.namenode.NNThroughputBenchmark.main(NNThroughputBenchmark.java:1534) > ... > {code} > However, the root cause is different from HDFS-9387. > From ReplicationStats#generateInputs, *nameNode* is uninitialized before use, > which causes the NPE. > {code} > final FSNamesystem namesystem = nameNode.getNamesystem(); > {code} > From NNThroughputBenchmark#run, nameNode is only initialized when -namenode > option is not specified. The fix is to initialize it properly in the else > block when -namenode option is specified OR we should block this if it is not > supported. > {code} > if (namenodeUri == null) { > nameNode = NameNode.createNameNode(argv, config); > NamenodeProtocols nnProtos = nameNode.getRpcServer(); > nameNodeProto = nnProtos; > clientProto = nnProtos; > dataNodeProto = nnProtos; > refreshUserMappingsProto = nnProtos; > bpid = nameNode.getNamesystem().getBlockPoolId(); > } else { > FileSystem.setDefaultUri(getConf(), namenodeUri); > DistributedFileSystem dfs = (DistributedFileSystem) > FileSystem.get(getConf()); > final URI nnUri = new URI(namenodeUri); > nameNodeProto = DFSTestUtil.getNamenodeProtocolProxy(config, nnUri, > UserGroupInformation.getCurrentUser()); > > {code} -- This message was sent by Atlassian JIRA (v6.3.4#6332)