Jing Zhao created HDFS-4093: ------------------------------- Summary: In branch-1-win, AzureBlockPlacementPolicy#chooseTarget only returns one DN when replication factor is greater than 3. Key: HDFS-4093 URL: https://issues.apache.org/jira/browse/HDFS-4093 Project: Hadoop HDFS Issue Type: Bug Reporter: Jing Zhao Assignee: Jing Zhao
In branch-1-win, when AzureBlockPlacementPolicy (which extends the BlockPlacementPolicyDefault) is used, if the client increases the number of replicas (e.g., from 3 to 10), AzureBlockPlacementPolicy#chooseTarget will return only 1 Datanode each time. Thus in FSNameSystem#computeReplicationWorkForBlock, it is possible that the replication monitor may choose a datanode that has been chosen as target but still in the pendingReplications (because computeReplicationWorkForBlock does not check the pending replication before doing the chooseTarget). To avoid this "hit-the-same-datanode" scenario, we modify the AzureBlockPlacementPolicy#chooseTarget to make it return multiple DN. -- 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