Skip to content

SPY-1737: avoid excessive/redundant calls to DistributedFileSystem.isDirectory #227

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Apr 20, 2018
Merged

SPY-1737: avoid excessive/redundant calls to DistributedFileSystem.isDirectory #227

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ebada58
SPY-1737: avoid excessive/redundant calls to DistributedFileSystem.is…
ianlcsd Apr 19, 2018
ce3a517
debug level logging
ianlcsd Apr 19, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 12 additions & 7 deletions sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveMetastoreCatalog.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -227,23 +227,28 @@ private[hive] class HiveMetastoreCatalog(sparkSession: SparkSession) extends Log
result.copy(output = newOutput)
}

/**
* Selecting directories at driver side by interacting with hdfs won't scale.
* TODO: migrate to parquet partitioning
*/
private[hive] def selectParquetLocationDirectories(
tableName: String,
locationOpt: Option[Path]): Seq[Path] = {
val start = System.currentTimeMillis
val hadoopConf = sparkSession.sparkContext.hadoopConfiguration
val paths: Option[Seq[Path]] = for {
selector <- sparkSession.sharedState.externalCatalog.findHadoopFileSelector
location <- locationOpt
fs = location.getFileSystem(hadoopConf)
selectedPaths <- selector.selectFiles(tableName, fs, location)
selectedDir = for {
selectedPath <- selectedPaths
if selectedPath
.getFileSystem(hadoopConf)
.isDirectory(selectedPath)
} yield selectedPath
// Csd's HadoopFileSelector should guarantee to return directories only,
selectedDir <- selector.selectFiles(tableName, fs, location)
if selectedDir.nonEmpty
} yield selectedDir
logDebug(
s"process duration of HiveMetastoreCatalog.selectParquetLocationDirectories(" +
s"$tableName, $locationOpt): ${System.currentTimeMillis - start}, selected directories: " +
s"${paths.map(_.size).getOrElse(0)}")

paths.getOrElse(Seq(locationOpt.orNull))
}

Expand Down
8 changes: 0 additions & 8 deletions sql/hive/src/test/scala/org/apache/spark/sql/hive/HiveMetastoreCatalogSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -223,14 +223,6 @@ class ParquetLocationSelectionSuite extends QueryTest with SQLTestUtils with Tes
hmc.selectParquetLocationDirectories("sometable", Option(new Path("somewhere")))
}

// ensure file existence for somewhere/sometable
somewhereSometable.delete()
somewhereSometable.createNewFile()
// somewhere/sometable is a file => will not be selected
assertResult(Seq(new Path("somewhere"))) {
hmc.selectParquetLocationDirectories("otherplace", Option(new Path("somewhere")))
}

// no location specified, none selected
assertResult(Seq(null)) {
hmc.selectParquetLocationDirectories("sometable", Option(null))
Expand Down