[ https://issues.apache.org/jira/browse/HDFS-7310?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14200881#comment-14200881 ]
Jing Zhao commented on HDFS-7310: --------------------------------- Thanks for working on this, [~vinayrpet]! The patch looks good to me. Only some minor comments: # Maybe we can let DatanodeStorageInfo#addBlock return an enum ? # Replicas in ram_disk currently has its own lazy-persist mechanism. Thus for the new moveBlockAcrossStorage method, maybe we can simply ignore the replica (or throw exception) if it's currently on transient storage? # The patch already has a new unit test in TestMover. Can we also have some tests only focusing on the new {{moveBlockAcrossStorage}} method and the {{replaceBlock}} protocol? > Mover can give first priority to local DN if it has target storage type > available in local DN > --------------------------------------------------------------------------------------------- > > Key: HDFS-7310 > URL: https://issues.apache.org/jira/browse/HDFS-7310 > Project: Hadoop HDFS > Issue Type: Improvement > Components: balancer & mover > Affects Versions: 3.0.0 > Reporter: Uma Maheswara Rao G > Assignee: Vinayakumar B > Attachments: HDFS-7310-001.patch, HDFS-7310-002.patch > > > Currently Mover logic may move blocks to any DN which had target storage > type. But if the src DN has target storage type then mover can give highest > priority to local DN. If local DN does not contains target storage type, then > it can assign to any DN as the current logic does. > This is a thought, have not go through the code fully yet. > Thoughts? -- This message was sent by Atlassian JIRA (v6.3.4#6332)