[
https://issues.apache.org/jira/browse/HDFS-5207?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13768113#comment-13768113
]
Tsz Wo (Nicholas), SZE commented on HDFS-5207:
----------------------------------------------
{code}
private DatanodeDescriptor chooseTarget(int numOfReplicas,
- DatanodeDescriptor writer,
- Map<Node, Node> excludedNodes,
+ Node writer,
+ Set<Node> excludedNodes,
long blocksize,
int maxNodesPerRack,
List<DatanodeDescriptor> results,
final boolean avoidStaleNodes) {
if (numOfReplicas == 0 || clusterMap.getNumOfLeaves()==0) {
- return writer;
+ return null;
}
int totalReplicasExpected = numOfReplicas + results.size();
int numOfResults = results.size();
boolean newBlock = (numOfResults==0);
- if (writer == null && !newBlock) {
+ if ((writer == null || !(writer instanceof DatanodeDescriptor)) &&
!newBlock) {
writer = results.get(0);
}
...
}
{code}
- In the above method, chooseTarget(..) is supposed to return the local node.
-* It should return writer instead of null even if (numOfReplicas == 0 ||
clusterMap.getNumOfLeaves()==0).
-* We need to change the return type to Node since the local node could
theoretically be a non-DatanodeDescriptor. Then, we don't need to cast writer
to DatanodeDescriptor when returning it.
-* You may also want to add a @return tag in the javadoc to mention that the
method returns the local node but not the chosen node.
> In BlockPlacementPolicy, update 2 parameters of chooseTarget()
> --------------------------------------------------------------
>
> Key: HDFS-5207
> URL: https://issues.apache.org/jira/browse/HDFS-5207
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: namenode
> Reporter: Junping Du
> Assignee: Junping Du
> Attachments: HDFS-5207-v1.patch, HDFS-5207-v2.patch
>
>
> We should change chooseTarget(..., DatanodeDescriptor writer, ...) to
> chooseTarget (..., node writer, ...) as the only important property of writer
> is to identify other nodes' location relationship so more generic one could
> be better. It also helps to cover cases that client node is not a Datanode.
> Also, the excluded nodes are not necessary to be a HashMap with the same key
> and value, Set is more proper in this case.
--
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
