[
https://issues.apache.org/jira/browse/HIVE-10031?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14375328#comment-14375328
]
Ferdinand Xu commented on HIVE-10031:
-------------------------------------
Hi [~jxiang], can you take a look at this patch?
Do we need to copy the old conf in the following code? The jobConf has already
be cloned in getSplit method when trying to push filter.
<noformat> jobConf = projectionPusher.pushProjectionsAndFilters(conf,
finalPath.getParent()); </noformat>
<noformat>
if (skipTimestampConversion ^ HiveConf.getBoolVar(
conf, HiveConf.ConfVars.HIVE_PARQUET_TIMESTAMP_SKIP_CONVERSION)) {
conf = new JobConf(oldJobConf);
HiveConf.setBoolVar(conf,
HiveConf.ConfVars.HIVE_PARQUET_TIMESTAMP_SKIP_CONVERSION,
skipTimestampConversion);
}
</noformat>
> Modify the using of jobConf variable in ParquetRecordReaderWrapper constructor
> ------------------------------------------------------------------------------
>
> Key: HIVE-10031
> URL: https://issues.apache.org/jira/browse/HIVE-10031
> Project: Hive
> Issue Type: Sub-task
> Reporter: Dong Chen
> Assignee: Dong Chen
> Attachments: HIVE-10031-parquet.patch
>
>
> In {{ParquetRecordReaderWrapper}} constructor, it create splits, set
> projections and filters in conf, create task context, and then create Parquet
> record reader. In this procedure, we could improve the logic of conf usage:
> 1. the clone of jobConf is not necessary. This could speed up getRecordReader
> a little.
> 2. the updated jobConf is not passed to Parquet in one case.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
