[ https://issues.apache.org/jira/browse/HADOOP-8928?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13503348#comment-13503348 ]
liang xie commented on HADOOP-8928: ----------------------------------- After applied the patch, we can update the rack config script or file, then issue "hdfs dfsadmin -refreshTopology", then startup the new datanode, it'll register to namenode successfully, w/o to restart namenode:) I tested both ScriptBasedMapping and TableMapping impl, all is OK There's a nit for "load()" in new TableMapping class, it'll be invoked many times while cluster starting or refreshTopology, but it's not a big issue, cause the OS page cache will reduce the physical IO ops while cluster starting. > Add ability to reset topologies on master nodes > ----------------------------------------------- > > Key: HADOOP-8928 > URL: https://issues.apache.org/jira/browse/HADOOP-8928 > Project: Hadoop Common > Issue Type: Improvement > Components: net > Affects Versions: 3.0.0, 2.0.2-alpha > Reporter: Shinichi Yamashita > Attachments: HADOOP-8928.txt > > > For a topology decision of DataNode and TaskTracker, ScriptBasedMapping > (probably TableMapping) confirms HashMap first. > To decide topology of DataNode and TaskTracker again, it is necessary to > restart NameNode and JobTracker. > Therefore, it is necessary to change (or clear) HashMap function without > restarting NameNode and JobTracker. -- 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