[
https://issues.apache.org/jira/browse/TEZ-3849?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16205241#comment-16205241
]
Kuhu Shukla edited comment on TEZ-3849 at 10/16/17 5:59 PM:
------------------------------------------------------------
Thank you for the patch [~jacobtolar]!
I am fine with peekPartition logic since we don't advance the kvIndex and still
need to check if the data for that partition exists or not.
Small comments on the patch:
1. Do we need the pom.xml changes? Ideally we should not add dependencies in
runtime-library for mapreduce client-core package.
2. Adding the verify counters to the test : {{testWithCombiner}} at Line 461
would be nice.
3. Also, we can make DummyCombiner an inner class to TestPipelineSorter.
was (Author: kshukla):
Thank you for the patch [~jacobtolar]!
I am fine with peekPartition logic since we don't advance the kvIndex and still
need to check if the data for that partition exists or not.
Small comments on the patch:
1. Do we need the pom.xml changes? Ideally we should not add dependencies in
runtime-library for mapreduce client-core package.
2. Adding the verify counters to the test : {{testWithCombiner}} at Line 461
would be nice.
> Combiner+PipelinedSorter silently drops records
> -----------------------------------------------
>
> Key: TEZ-3849
> URL: https://issues.apache.org/jira/browse/TEZ-3849
> Project: Apache Tez
> Issue Type: Bug
> Reporter: Jacob Tolar
> Assignee: Jacob Tolar
> Attachments: TEZ-3849.1.patch, TEZ-3849.2.patch, TEZ-3849.3.patch,
> TEZ-3849.4.patch
>
>
> This bug was introduced in
> https://github.com/apache/tez/commit/a47e8fcbea5eeab5a7cf812271d329524cc02dba?diff=split
>
> when combiner != null, the change in this commit passes kvIter with next()
> having already been called. This ends up (silently) dropping the first record
> in the partition.
> Will submit PR and attach patch. [~jeagles], not sure if this is the way you
> want to fix or not but it does fix my tests.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
