hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Jing Zhao (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (HDFS-9866) BlockManager#chooseExcessReplicasStriped may weaken rack fault tolerance
Date Sat, 27 Feb 2016 02:10:18 GMT

     [ https://issues.apache.org/jira/browse/HDFS-9866?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Jing Zhao updated HDFS-9866:
----------------------------
    Attachment: HDFS-9866.000.patch

Update a patch to fix. For a striped block, chooseReplicasToDelete should distinguish 1) available
replicas and 2) candidates for deletion. The former is used to split the deletion candidates
into the {{exactlyOne}} and {{moreThanOne}} sets.

> BlockManager#chooseExcessReplicasStriped may weaken rack fault tolerance
> ------------------------------------------------------------------------
>
>                 Key: HDFS-9866
>                 URL: https://issues.apache.org/jira/browse/HDFS-9866
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: namenode
>    Affects Versions: 3.0.0
>            Reporter: Takuya Fukudome
>            Assignee: Jing Zhao
>         Attachments: HDFS-9866.000.patch
>
>
> In [~tfukudom]'s system tests, we find the following issue:
> A striped block group B has redundant internal block replicas. 9 internal blocks are
stored in 10 datanodes across 6 racks. Datanode d1 and d2 both store a replica for internal
block b1. d1's rack contains multiple internal blocks while d2's rack only has b1. Then when
choosing a duplicated replica to delete,  the current implementation may wrongly choose d2
thus causes the total number of racks to be decreased to 5.



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

Mime
View raw message