[
https://issues.apache.org/jira/browse/HIVE-10213?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14529452#comment-14529452
]
Sushanth Sowmyan commented on HIVE-10213:
-----------------------------------------
This patch set off some warning flags for me with regards to the traditional
M-R usecase, but it's because it's been a while since I looked at this piece of
code. The traditional M-R usecase is still fine, because the
DynamicPartitionFileRecordWriterContainer.close() will register an appropriate
TaskCommitterProxy, and a commit on the OutputCommitter will be called in the
same process scope, thus making it okay. For pig-based optimizations also, it'd
continue to be okay as the singleton retains it in memory.
+1, and I'm okay with committing this patch as-is, tests have already run on
this, and this section of code has not changed since then.
> MapReduce jobs using dynamic-partitioning fail on commit.
> ---------------------------------------------------------
>
> Key: HIVE-10213
> URL: https://issues.apache.org/jira/browse/HIVE-10213
> Project: Hive
> Issue Type: Bug
> Components: HCatalog
> Reporter: Mithun Radhakrishnan
> Assignee: Mithun Radhakrishnan
> Attachments: HIVE-10213.1.patch
>
>
> I recently ran into a problem in {{TaskCommitContextRegistry}}, when using
> dynamic-partitions.
> Consider a MapReduce program that reads HCatRecords from a table (using
> HCatInputFormat), and then writes to another table (with identical schema),
> using HCatOutputFormat. The Map-task fails with the following exception:
> {code}
> Error: java.io.IOException: No callback registered for
> TaskAttemptID:attempt_1426589008676_509707_m_000000_0@hdfs://crystalmyth.myth.net:8020/user/mithunr/mythdb/target/_DYN0.6784154320609959/grid=__HIVE_DEFAULT_PARTITION__/dt=__HIVE_DEFAULT_PARTITION__
> at
> org.apache.hive.hcatalog.mapreduce.TaskCommitContextRegistry.commitTask(TaskCommitContextRegistry.java:56)
> at
> org.apache.hive.hcatalog.mapreduce.FileOutputCommitterContainer.commitTask(FileOutputCommitterContainer.java:139)
> at org.apache.hadoop.mapred.Task.commit(Task.java:1163)
> at org.apache.hadoop.mapred.Task.done(Task.java:1025)
> at org.apache.hadoop.mapred.MapTask.run(MapTask.java:345)
> at org.apache.hadoop.mapred.YarnChild$2.run(YarnChild.java:163)
> at java.security.AccessController.doPrivileged(Native Method)
> at javax.security.auth.Subject.doAs(Subject.java:415)
> at
> org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1694)
> at org.apache.hadoop.mapred.YarnChild.main(YarnChild.java:158)
> {code}
> {{TaskCommitContextRegistry::commitTask()}} uses call-backs registered from
> {{DynamicPartitionFileRecordWriter}}. But in case {{HCatInputFormat}} and
> {{HCatOutputFormat}} are both used in the same job, the
> {{DynamicPartitionFileRecordWriter}} might only be exercised in the Reducer.
> I'm relaxing the IOException, and log a warning message instead of just
> failing.
> (I'll post the fix shortly.)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
