[ https://issues.apache.org/jira/browse/HDFS-3902?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13450768#comment-13450768 ]
Andy Isaacson commented on HDFS-3902: ------------------------------------- Since HDFS-3828 fixed the block scanner to not repeatedly rescan small blockpools, the following test in TestDatanodeBlockScanner times out after 13 minutes in {{waitReplication}}: {code} 207 assertTrue(MiniDFSCluster.corruptReplica(0, block)); 208 assertTrue(MiniDFSCluster.corruptReplica(1, block)); 209 assertTrue(MiniDFSCluster.corruptReplica(2, block)); ... 219 // We now have the blocks to be marked as corrupt and we get back all 220 // its replicas 221 DFSTestUtil.waitReplication(fs, file1, (short)3); 222 assertTrue(DFSTestUtil.allBlockReplicasCorrupt(cluster, file1, 0)); {code} That assertion seems very odd to me: it claims that a replication=3 block with two corrupt replicas should report 1 location, but once we corrupt the third replica the block should report all three (corrupt!) locations. While validating this test regression, I found that TestDatanodeBlockScanner varies its completion time between 1.5 and 12 minutes when run under {{mvn test}}. > TestDatanodeBlockScanner is flaky, broke entirely after HDFS-3828 > ----------------------------------------------------------------- > > Key: HDFS-3902 > URL: https://issues.apache.org/jira/browse/HDFS-3902 > Project: Hadoop HDFS > Issue Type: Bug > Affects Versions: 2.1.0-alpha > Reporter: Andy Isaacson > Assignee: Andy Isaacson > Priority: Minor > > Since HDFS-3828 fixed the block scanner to not repeatedly rescan small > blockpools, TestDatanodeBlockScanner times out after 13 minutes in > {{waitReplication. -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira