Skip to content

[SPARK-15269][SQL] Removes unexpected empty table directories created while creating external Spark SQL data sourcet tables. #13270

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

Closed
liancheng wants to merge 8 commits into from
Closed

[SPARK-15269][SQL] Removes unexpected empty table directories created while creating external Spark SQL data sourcet tables. #13270

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
528b835
Fixes SPARK-15269
liancheng May 24, 2016
fa7b5b6
Fixes test failures
liancheng May 24, 2016
0aedf7b
Fixes more test failures
liancheng May 25, 2016
505f3f0
Handles Hive compatible tables properly
liancheng May 25, 2016
04af79d
Fixes newly added test case
liancheng May 25, 2016
6241289
More comments
liancheng May 25, 2016
336fb55
Moves changes in SessionCatalog to HiveExternalCatalog
liancheng May 31, 2016
7d0122f
Reverts minor refactoring that is not supported by Scala 2.10
liancheng May 31, 2016
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
Prev Previous commit
Next Next commit
Fixes test failures
  • Loading branch information
@liancheng
liancheng committed May 31, 2016
commit fa7b5b6f676553757da96ccd40944ec349a489af
29 changes: 21 additions & 8 deletions sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/catalog/SessionCatalog.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,24 +212,37 @@ class SessionCatalog(
* If no such database is specified, create it in the current database.
*/
def createTable(tableDefinition: CatalogTable, ignoreIfExists: Boolean): Unit = {
val db = formatDatabaseName(tableDefinition.identifier.database.getOrElse(getCurrentDatabase))
val table = formatTableName(tableDefinition.identifier.table)
val tableId = tableDefinition.identifier
val db = formatDatabaseName(tableId.database.getOrElse(getCurrentDatabase))
val table = formatTableName(tableId.table)
val newTableDefinition = tableDefinition.copy(identifier = TableIdentifier(table, Some(db)))
requireDbExists(db)

if (newTableDefinition.tableType == CatalogTableType.EXTERNAL) {
if (
// If this is an external data source table
tableDefinition.properties.contains("spark.sql.sources.provider") &&
newTableDefinition.tableType == CatalogTableType.EXTERNAL
) {
// !! HACK ALERT !!
//
// See https://issues.apache.org/jira/browse/SPARK-15269 for more details about why we have to
// set `locationUri` and then remove the directory after creating the external table:
val tablePath = defaultTablePath(newTableDefinition.identifier)
// Due to a restriction of Hive metastore, here we have to set `locationUri` to a temporary
// directory that doesn't exist yet but can definitely be successfully created, and then
// delete it right after creating the external data source table. This location will be
// persisted to Hive metastore as standard Hive table location URI, but Spark SQL doesn't
// really use it. Also, since we only do this workaround for external tables, deleting the
// directory after the fact doesn't do any harm.
//
// Please refer to https://issues.apache.org/jira/browse/SPARK-15269 for more details.

val tempPath = new Path(defaultTablePath(tableId), "-__PLACEHOLDER__").toString

try {
externalCatalog.createTable(
db,
newTableDefinition.withNewStorage(locationUri = Some(tablePath)),
newTableDefinition.withNewStorage(locationUri = Some(tempPath)),
ignoreIfExists)
} finally {
val path = new Path(tablePath)
val path = new Path(tempPath)
FileSystem.get(path.toUri, hadoopConf).delete(path, true)
}
} else {
Expand Down
6 changes: 3 additions & 3 deletions sql/core/src/main/scala/org/apache/spark/sql/SparkSession.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -556,9 +556,9 @@ class SparkSession private(
}


/* ------------------------ *
| Catalog-related methods |
* ----------------- ------ */
/* ------------------------- *
| Catalog-related methods |
* ------------------------- */

/**
* Interface through which the user may create, drop, alter or query underlying
Expand Down
5 changes: 4 additions & 1 deletion sql/core/src/main/scala/org/apache/spark/sql/execution/command/createDataSourceTables.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -400,7 +400,10 @@ object CreateDataSourceTableUtils extends Logging {
schema = relation.schema.map { f =>
CatalogColumn(f.name, f.dataType.catalogString)
},
properties = tableProperties.toMap,
// Removes the provider property since we are gonna saving this table as a Hive compatible
// one, and other places use this property to check whether a table is a data source table
// (e.g. `DDLUtils.isDatasourceTable`).
properties = (tableProperties - "spark.sql.sources.provider").toMap,
Copy link
Contributor

@xwu0226 xwu0226 May 24, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see. So the hive compatible datasource table will not be resolved as datasource relation anymore. Correct?

Copy link
Contributor Author

@liancheng liancheng May 25, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Correct.

Copy link
Contributor Author

@liancheng liancheng May 25, 2016 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Unfortunately this doesn't work according to the last Jenkins build failure, because we still want to recognize this table as data source table when saving data into the table using CreateDataSourceTableAsSelectCommand. My last commit tries to fix this by checking .storage.locationUri.isEmpty. This should work because data source tables never set this field.

viewText = None)
}

Expand Down
2 changes: 1 addition & 1 deletion sql/hive/src/main/scala/org/apache/spark/sql/hive/client/HiveClientImpl.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -377,7 +377,7 @@ private[hive] class HiveClientImpl(
// allows directory paths as location URIs while Spark SQL data source tables also
// allows file paths. So the standard Hive `dataLocation` is meaningless for Spark SQL
// data source tables.
DDLUtils.isDatasourceTable(properties)
DDLUtils.isDatasourceTable(properties) && h.getTableType == HiveTableType.EXTERNAL_TABLE
},
Copy link
Contributor

@yhuai yhuai May 31, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Why do we need this check?

Copy link
Contributor Author

@liancheng liancheng May 31, 2016

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Because we have to store the placeholder location URI into metastore for external data source tables, and I'd like to avoid exposing it to user space.

Copy link
Contributor Author

@liancheng liancheng May 31, 2016 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Technically, it doesn't harm to keep this field since Spark SQL doesn't use it. But this placeholder location URI doesn't make sense anywhere, and it can be error prone to keep it in the locationUri field since future maintainers may think it is the real data location.

inputFormat = Option(h.getInputFormatClass).map(_.getName),
outputFormat = Option(h.getOutputFormatClass).map(_.getName),
Expand Down