[
https://issues.apache.org/jira/browse/HDFS-15655?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17236121#comment-17236121
]
Ayush Saxena commented on HDFS-15655:
-------------------------------------
{quote}. But I am interested what exception dfsclient meets when move related
blocks.
{quote}
I couldn't check the code changes here, but Client I don't think would differ
b/w cold or hot blocks, if the stream is reading. In case the blocks are moved,
IIRC it would first try all three replicas and if all three seems to be not
available it will go and refetch the blocks again from the namenode and retry a
couple of times. Something like this...
I think the intention here could be that cold blocks would be having no or low
load and this refetching would be less as compared to HOT blocks.
> Add option to make balancer prefer to get cold blocks
> -----------------------------------------------------
>
> Key: HDFS-15655
> URL: https://issues.apache.org/jira/browse/HDFS-15655
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: balancer & mover
> Reporter: Yang Yun
> Assignee: Yang Yun
> Priority: Minor
> Attachments: HDFS-15655.001.patch, HDFS-15655.002.patch,
> HDFS-15655.003.patch, HDFS-15655.004.patch
>
>
> We met two issues when using balancer.
> # Moving hot files may cause failing of dfsclient reading.
> # Some blocks of temporary files are moved and they are deleted soon.
> Add a config key " dfs.namenode.hot.block.time.interval", the balancer
> prefer to get the blocks which are belong to the cold files created before
> this time period.
> Also add a option "-hotBlockTimeInterval" to the command line of balancer for
> inputting.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-h...@hadoop.apache.org
