[
https://issues.apache.org/jira/browse/HIVE-1608?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15139432#comment-15139432
]
Chaoyu Tang commented on HIVE-1608:
-----------------------------------
INSERT OVERWRITE [Local] DIRECTORY is actually not affected by this change, and
I have had some tests and verified it. It is because Hive already uses the
default tabledesc whose fileformat is hardcoded as "TextFile" for these cases.
See related code:
{code}
SemanticAnalyzer.java -- line 6523:
if (qb.getIsQuery()) {
String fileFormat = HiveConf.getVar(conf,
HiveConf.ConfVars.HIVEQUERYRESULTFILEFORMAT);
table_desc = PlanUtils.getDefaultQueryOutputTableDesc(cols, colTypes,
fileFormat);
} else {
table_desc = PlanUtils.getDefaultTableDesc(qb.getDirectoryDesc(),
cols, colTypes);
}
---
PlanUtils.java -- 211, 224
public static TableDesc getDefaultTableDesc(String separatorCode,
String columns, String columnTypes, boolean lastColumnTakesRestOfTheLine)
{
return getTableDesc(LazySimpleSerDe.class, separatorCode, columns,
columnTypes, lastColumnTakesRestOfTheLine);
}
public static TableDesc getTableDesc(
Class<? extends Deserializer> serdeClass, String separatorCode,
String columns, String columnTypes, boolean lastColumnTakesRestOfTheLine,
boolean useDelimitedJSON) {
return getTableDesc(serdeClass, separatorCode, columns, columnTypes,
lastColumnTakesRestOfTheLine, useDelimitedJSON, "TextFile");
}
{code}
> use sequencefile as the default for storing intermediate results
> ----------------------------------------------------------------
>
> Key: HIVE-1608
> URL: https://issues.apache.org/jira/browse/HIVE-1608
> Project: Hive
> Issue Type: Bug
> Components: Query Processor
> Affects Versions: 0.7.0
> Reporter: Namit Jain
> Assignee: Brock Noland
> Attachments: HIVE-1608.1.patch, HIVE-1608.2.patch, HIVE-1608.3.patch,
> HIVE-1608.4.patch, HIVE-1608.patch
>
>
> The only argument for having a text file for storing intermediate results
> seems to be better debuggability.
> But, tailing a sequence file is possible, and it should be more space
> efficient
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
