[
https://issues.apache.org/jira/browse/HIVE-7776?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Chengxiang Li updated HIVE-7776:
--------------------------------
Attachment: HIVE-7776.1-spark.patch
Hive get task Id through 2 ways in Utilities::getTaskId:
# get parameter value of mapred.task.id from configuration.
# generate random value while #1 return null.
Currently, Hive on Spark can't get parameter value of mapred.task.id from
configuration.
FileSinkOperator use taskid to distinct different bucket file name,
FileSinkOperator should take taskid as field variable and initiate it only once
since one FileSinkOperator instance only refered in one task. but
FileSinkOperator call Utilities::getTaskId to get new taskId each time, for
this issue, it would cause more bucket files than bucket number, which lead to
unexpected result of tablesample queries.
> enable sample10.q.[Spark Branch]
> --------------------------------
>
> Key: HIVE-7776
> URL: https://issues.apache.org/jira/browse/HIVE-7776
> Project: Hive
> Issue Type: Sub-task
> Components: Spark
> Reporter: Chengxiang Li
> Assignee: Chengxiang Li
> Attachments: HIVE-7776.1-spark.patch
>
>
> sample10.q contain dynamic partition operation, should enable this qtest
> after hive on spark support dynamic partition.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
