[
https://issues.apache.org/jira/browse/HADOOP-4322?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12855483#action_12855483
]
Vinod K V commented on HADOOP-4322:
-----------------------------------
When this patch is tested internally, we found some problem - the job gives out
negative "input byes" count when using this outputformat. I don't have any
context about this, but just updating this bug with the analysis by Rahul..
{code}
ObjectFileRecordReader has getPos() method implementation , this method is
giving incorrect values for offset.
Code flow in the framework is like below.
=====
beforePos = getPos();
//call to user's record reader 'next() method.
afterPos = getPos();
//then for counter we do the following:
inputByteCounter.increment(afterPos - beforePos);//this is the counter which is
//in question
=====
(ObjectFileRecordReader's getPos() method ) afterPos < beforePos , this is
resulting in the -ve increment to the counter.
{code}
So this patch shouldn't be committed as is without a relook.
> Input/Output Format for TFile
> -----------------------------
>
> Key: HADOOP-4322
> URL: https://issues.apache.org/jira/browse/HADOOP-4322
> Project: Hadoop Common
> Issue Type: New Feature
> Reporter: Amir Youssefi
> Assignee: Amir Youssefi
> Attachments: ObjectFileInputOutputFormat_1.patch
>
>
> Input/Output Format for TFile
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
