hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Kihwal Lee (JIRA)" <j...@apache.org>
Subject [jira] [Comment Edited] (HDFS-9620) Slow writer may fail permanently if pipeline breaks.
Date Mon, 11 Jan 2016 19:00:41 GMT

    [ https://issues.apache.org/jira/browse/HDFS-9620?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15092485#comment-15092485
] 

Kihwal Lee edited comment on HDFS-9620 at 1/11/16 7:00 PM:
-----------------------------------------------------------

For readers, a new block token is acquired by calling {{getBlockLocations()}}. Writers get
a block from {{addBlock()}} and it cannot be called multiple times for the same block.  Can
we make a change to {{addBlock()}} to support it while maintaining compatibility? Or will
it be better to add a new method to the protocol?


was (Author: kihwal):
For readers, a new block token is acquired by calling {{getBlockLocations()}}. Writers get
a block from {{addBlock()}} and it cannot be called multiple times.  Can we make a change
to {{addBlock()}} to support it while maintaining compatibility? Or will it be better to add
a new method to the protocol?

> Slow writer may fail permanently if pipeline breaks.
> ----------------------------------------------------
>
>                 Key: HDFS-9620
>                 URL: https://issues.apache.org/jira/browse/HDFS-9620
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: hdfs-client, security
>            Reporter: Kihwal Lee
>            Priority: Critical
>
> During a block write to a datanode, if the block write time exceed the block token expiration,
the client will not be able to reestablish a block output stream. E.g. if a node in the pipeline
dies, the pipeline recovery won't work.



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

Mime
View raw message