hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Li Bo (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-8839) Erasure Coding: client occasionally gets less block locations when some datanodes fail
Date Mon, 03 Aug 2015 05:31:04 GMT

    [ https://issues.apache.org/jira/browse/HDFS-8839?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14651460#comment-14651460
] 

Li Bo commented on HDFS-8839:
-----------------------------

hi, Zhe, I remember HDFS-8820 handles the situation that a cluster is smaller than 9 datanodes.
For a cluster with 9 nodes, if one node fails, the namenode should still allocate 9 locations
even it knows one of them is invalid. Otherwise, the client writing can't go on.

> Erasure Coding: client occasionally gets less block locations when some datanodes fail

> ---------------------------------------------------------------------------------------
>
>                 Key: HDFS-8839
>                 URL: https://issues.apache.org/jira/browse/HDFS-8839
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>            Reporter: Li Bo
>            Assignee: Li Bo
>
> 9 datanodes, write two block groups. A datanode dies when writing the first block group.
When client retrieves the second block group from namenode, the returned block group only
contains 8 locations occasionally.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message