Details
-
Improvement
-
Status: Patch Available
-
Major
-
Resolution: Unresolved
-
2.8.2, 3.0.0
-
None
-
None
Description
When copying large amount of data from one cluster to another via Distcp, and the Distcp jobs run in the target cluster, the datanode local usage would be imbalanced. Because the default placement policy chooses the local node to store the first replication.
In https://issues.apache.org/jira/browse/HDFS-3702 we add a flag in DFSClient to avoid replicating to the local datanode. We can make use of this flag in Distcp.
Attachments
Attachments
Issue Links
- links to