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] [Commented] (HDFS-7672) Erasure Coding: handle write failure for stripping coding blocks
Date Mon, 04 May 2015 23:56:06 GMT

    [ https://issues.apache.org/jira/browse/HDFS-7672?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14527582#comment-14527582
] 

Jing Zhao commented on HDFS-7672:
---------------------------------

Thanks for working on this, Nicholas! Some comments on the current patch:
# In the current patch we do not have the logic about bumping the GenerationStamp when hitting
a failure. I think we can do it either in this jira or in a separate one.
# The current patch only captures failure while putting packets into queues and cannot guarantee
to catch failures while writing packets out to DN.
# We should still keep "volatile" for {{streamerClosed}} since it is accessed out of lock
protection in run().
# We do not need to add length to a parity block if its writing fails
# Can we move {{writeParity}} out of {{CellBuffers}} and make {{CellBuffers}} only about the
data/checksum buffers (i.e., no extra outputstream logic)?
# Maybe we can print all the datanodes in a separate method called {{getNodeListString}} instead
of {{toString}}?
# Need to fix the javadoc for {{DFSStripedOutputStream#generatePackets}}: buffer -> byteBuffer
# javadoc for CellBuffers?
# The two for loop in {{encode}} method can be replaced by System.arrayCopy.
# The variable {{hasEndedBlocks}} may need to be renamed. Some javadoc will also be helpful
for understanding.



> Erasure Coding: handle write failure for stripping coding blocks
> ----------------------------------------------------------------
>
>                 Key: HDFS-7672
>                 URL: https://issues.apache.org/jira/browse/HDFS-7672
>             Project: Hadoop HDFS
>          Issue Type: Sub-task
>          Components: hdfs-client
>            Reporter: Tsz Wo Nicholas Sze
>            Assignee: Tsz Wo Nicholas Sze
>         Attachments: h7672_20150504.patch, h7672_20150504b.patch
>
>
> In *stripping* case, for (6, 3)-Reed-Solomon, a client writes to 6 data blocks and 3
parity blocks concurrently.  We need to handle datanode or network failures when writing a
EC BlockGroup.



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

Mime
View raw message