[
https://issues.apache.org/jira/browse/TEZ-1148?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14031739#comment-14031739
]
Mariappan Asokan commented on TEZ-1148:
---------------------------------------
Hitesh,
Thanks for your feedback. On the reduce side, there is only one way to
specify the old reducer. The current code does not have problem with that. On
the map side, user can specify either an old Mapper or an old MapRunnable
implementation. Currently, the code checks only for the presence of old
Mapper. This causes the problem mentioned.
I will add a test case to {{TestMapProcessor.java}} as part of a new version of
the patch.
> Exception thrown when old MapRunnable implementation is specified for a
> MapProcessor vertex
> -------------------------------------------------------------------------------------------
>
> Key: TEZ-1148
> URL: https://issues.apache.org/jira/browse/TEZ-1148
> Project: Apache Tez
> Issue Type: Bug
> Affects Versions: 0.4.0, 0.5.0
> Reporter: Mariappan Asokan
> Assignee: Mariappan Asokan
> Attachments: tez-1148.patch
>
>
> When a MapProcessor vertex was configured with an implementation of
> MapRunnable, following exception was thrown:
> {code}
> Exception in thread "main" java.io.IOException:
> mapreduce.job.inputformat.class is incompatible with map compatability mode.
> at
> org.apache.tez.mapreduce.hadoop.MRHelpers.ensureNotSet(MRHelpers.java:629)
> at
> org.apache.tez.mapreduce.hadoop.MRHelpers.setUseNewAPI(MRHelpers.java:603)
> ....
> {code}
> The problem seems to be in {{MRHelpers.setUseNewAPI().}} The check for the
> existence of a {{MapRunnable}} implementation is missing which leads to an
> incorrect conclusion that new APIs are used. This problem exists in the
> trunk version as well.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
