[
https://issues.apache.org/jira/browse/HADOOP-3863?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12618549#action_12618549
]
Tsz Wo (Nicholas), SZE commented on HADOOP-3863:
------------------------------------------------
According to CharsetDecoder javadoc
http://java.sun.com/javase/6/docs/api/java/nio/charset/CharsetDecoder.html,
there is a specific sequence of *decoding operations*. In particular, we
should call reset() before using the decoder. It seems we are not following it.
> Use a thread-local rather than static ENCODER/DECODER variables in Text for
> synchronization
> -------------------------------------------------------------------------------------------
>
> Key: HADOOP-3863
> URL: https://issues.apache.org/jira/browse/HADOOP-3863
> Project: Hadoop Core
> Issue Type: Bug
> Components: mapred
> Affects Versions: 0.18.0
> Reporter: Arun C Murthy
> Assignee: Arun C Murthy
> Priority: Critical
> Fix For: 0.19.0
>
> Attachments: HADOOP-3863_0_20080730.patch,
> HADOOP-3863_1_20080730.patch
>
>
> Text.{ENCODER|DECODER} are static variables which need to be synchronized in
> Text.{read|write}String; given that lots of RPC code calls
> Text.{read|write}String the contention for this lock shows up very
> significantly on large clusters.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
