[ https://issues.apache.org/jira/browse/HDFS-5114?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14118394#comment-14118394 ]
Zhe Zhang commented on HDFS-5114: --------------------------------- [~kihwal] Since this was created a year ago, do you happen to know if it has been resolved in the latest code? If not I'm happy to work on it. Thanks! > getMaxNodesPerRack() in BlockPlacementPolicyDefault does not take > decommissioning nodes into account. > ----------------------------------------------------------------------------------------------------- > > Key: HDFS-5114 > URL: https://issues.apache.org/jira/browse/HDFS-5114 > Project: Hadoop HDFS > Issue Type: Bug > Components: namenode > Affects Versions: 3.0.0, 2.1.0-beta > Reporter: Kihwal Lee > Assignee: Zhe Zhang > > If a large proportion of data nodes are being decommissioned, one or more > racks may not be writable. However this is not taken into account when the > default block placement policy module invokes getMaxNodesPerRack(). Some > blocks, especially the ones with a high replication factor, may not be able > to fully replicated until those nodes are taken out of dfs.include. It can > actually block decommissioning itself. -- This message was sent by Atlassian JIRA (v6.3.4#6332)