[
https://issues.apache.org/jira/browse/HDFS-10782?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16912952#comment-16912952
]
He Xiaoqiao commented on HDFS-10782:
------------------------------------
[~jojochuang] Thanks for picking this issue up,
[^HDFS-10782-branch-2.001.patch] is one demo patch based on branch-2.7, as you
said above, this patch implements with some issues indeed, memory footprint,
performance and so on, I would like to attach another new with unit test(yes,
it isn't hard to build one) based on branch trunk later. And welcome any
furthermore discussions. Thanks.
> Decrease memory frequent exchange of Centralized Cache Management when run
> balancer
> -----------------------------------------------------------------------------------
>
> Key: HDFS-10782
> URL: https://issues.apache.org/jira/browse/HDFS-10782
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: balancer & mover, caching
> Affects Versions: 2.7.1
> Reporter: He Xiaoqiao
> Assignee: He Xiaoqiao
> Priority: Major
> Labels: patch
> Attachments: HDFS-10782-branch-2.001.patch
>
>
> CachedBlocks currently are transparent for Balancer when active feature of
> centralized cache management. This makes DataNode exchange memory frequently,
> because Balancer does not Distinguish CachedBlock from blocks, so it may
> trigger mount of cached/uncached ops.
> I think namenode should avoid return CacheBlocks as much as possible when
> Balanacer#getblocks.
--
This message was sent by Atlassian Jira
(v8.3.2#803003)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
