[
https://issues.apache.org/jira/browse/HDFS-6840?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14113720#comment-14113720
]
Yi Liu commented on HDFS-6840:
------------------------------
Thanks [~andrew.wang], the patch looks good to me. Just one thing, we need
also remove the configuration from hdfs-default.xml
{code}
<property>
<name>dfs.namenode.randomize-block-locations-per-block</name>
<value>false</value>
<description>When fetching replica locations of a block, the replicas
are sorted based on network distance. This configuration parameter
determines whether the replicas at the same network distance are randomly
shuffled. By default, this is false, such that repeated requests for a
block's
replicas always result in the same order. This potentially improves page
cache
behavior. However, for some network topologies, it is desirable to shuffle
this
order for better load balancing.
</description>
</property>
{code}
> Clients are always sent to the same datanode when read is off rack
> ------------------------------------------------------------------
>
> Key: HDFS-6840
> URL: https://issues.apache.org/jira/browse/HDFS-6840
> Project: Hadoop HDFS
> Issue Type: Bug
> Affects Versions: 2.5.0
> Reporter: Jason Lowe
> Assignee: Andrew Wang
> Priority: Critical
> Attachments: hdfs-6840.001.patch
>
>
> After HDFS-6268 the sorting order of block locations is deterministic for a
> given block and locality level (e.g.: local, rack. off-rack), so off-rack
> clients all see the same datanode for the same block. This leads to very
> poor behavior in distributed cache localization and other scenarios where
> many clients all want the same block data at approximately the same time.
> The one datanode is crushed by the load while the other replicas only handle
> local and rack-local requests.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
