viirya commented on a change in pull request #27690:
URL: https://github.com/apache/spark/pull/27690#discussion_r446593989
##########
File path:
sql/hive/src/main/scala/org/apache/spark/sql/hive/execution/SaveAsHiveFile.scala
##########
@@ -97,12 +99,38 @@ private[hive] trait SaveAsHiveFile extends
DataWritingCommand {
options = Map.empty)
}
- protected def getExternalTmpPath(
+ // Mostly copied from Context.java#getMRTmpPath of Hive 2.3.
+ // Visible for testing.
+ private[execution] def getNonBlobTmpPath(
+ hadoopConf: Configuration,
+ sessionScratchDir: String,
+ scratchDir: String): Path = {
+
+ // Hive's getMRTmpPath uses nonLocalScratchPath + '-mr-10000',
+ // which is ruled by 'hive.exec.scratchdir' including file system.
+ // This is the same as Spark's #oldVersionExternalTempPath.
+ // Only difference between #oldVersionExternalTempPath and Hive 2.3.0's is
HIVE-7090.
+ // HIVE-7090 added user_name/session_id on top of 'hive.exec.scratchdir'
+ // Here it uses session_path unless it's emtpy, otherwise uses scratchDir.
+ val sessionPath = if (!sessionScratchDir.isEmpty) sessionScratchDir else
scratchDir
+ val mrScratchDir = oldVersionExternalTempPath(new Path(sessionPath),
hadoopConf, sessionPath)
Review comment:
When this new feature is enabled, it is possible that a scheme which
doesn't work for this feature is used, e.g. local scheme. If it is happened and
causes some error, end-users might not know how to deal with it.
Because we don't know if every scheme supports this feature, we use a list
of schemes as config value, instead of a boolean config. Similarly, I think we
should not reply on an assumption that `fs.default.name` always works for this
feature. Can we just restrict this feature to HDFS only?
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
