cloud-fan commented on a change in pull request #25330: [SPARK-28565][SQL]
DataFrameWriter saveAsTable support for V2 catalogs
URL: https://github.com/apache/spark/pull/25330#discussion_r311403624
##########
File path: sql/core/src/main/scala/org/apache/spark/sql/DataFrameWriter.scala
##########
@@ -485,7 +486,80 @@ final class DataFrameWriter[T] private[sql](ds:
Dataset[T]) {
* @since 1.4.0
*/
def saveAsTable(tableName: String): Unit = {
-
saveAsTable(df.sparkSession.sessionState.sqlParser.parseTableIdentifier(tableName))
+ import df.sparkSession.sessionState.analyzer.{AsTableIdentifier,
CatalogObjectIdentifier}
+ import org.apache.spark.sql.catalog.v2.CatalogV2Implicits._
+
+ val session = df.sparkSession
+ val useV1Sources =
+
session.sessionState.conf.useV1SourceWriterList.toLowerCase(Locale.ROOT).split(",")
+ val cls = DataSource.lookupDataSource(source, session.sessionState.conf)
+ val shouldUseV1Source = cls.newInstance() match {
+ case d: DataSourceRegister if useV1Sources.contains(d.shortName()) =>
true
+ case _ =>
useV1Sources.contains(cls.getCanonicalName.toLowerCase(Locale.ROOT))
+ }
+
+ val canUseV2 = !shouldUseV1Source &&
classOf[TableProvider].isAssignableFrom(cls)
+ val sessionCatalogOpt = session.sessionState.analyzer.sessionCatalog
+
+ session.sessionState.sqlParser.parseMultipartIdentifier(tableName) match {
+ case CatalogObjectIdentifier(Some(catalog), ident) =>
+ saveAsTable(catalog.asTableCatalog, ident, modeForDSV2)
+
+ case CatalogObjectIdentifier(None, ident) if canUseV2 &&
sessionCatalogOpt.isDefined =>
+ // We pass in the modeForDSV1, as using the V2 session catalog should
maintain compatibility
+ // for now.
+ saveAsTable(sessionCatalogOpt.get.asTableCatalog, ident, modeForDSV1)
+
+ case AsTableIdentifier(tableIdentifier) =>
+ saveAsTable(tableIdentifier)
+ }
+ }
+
+
+ private def saveAsTable(catalog: TableCatalog, ident: Identifier, mode:
SaveMode): Unit = {
+ val partitioning = partitioningColumns.map { colNames =>
+ colNames.map(name => IdentityTransform(FieldReference(name)))
+ }.getOrElse(Seq.empty[Transform])
+ val bucketing = bucketColumnNames.map { cols =>
+ Seq(BucketTransform(LiteralValue(numBuckets.get, IntegerType),
cols.map(FieldReference(_))))
+ }.getOrElse(Seq.empty[Transform])
+ val partitionTransforms = partitioning ++ bucketing
+
+ val tableOpt = try Option(catalog.loadTable(ident)) catch {
+ case _: NoSuchTableException => None
+ }
+
+ val command = (mode, tableOpt) match {
+ case (SaveMode.Append, Some(table)) =>
+ AppendData.byName(DataSourceV2Relation.create(table), df.logicalPlan)
+
+ case (SaveMode.Overwrite, _) =>
+ ReplaceTableAsSelect(
+ catalog,
+ ident,
+ partitionTransforms,
+ df.queryExecution.analyzed,
+ Map.empty, // properties can't be specified through this
API
+ extraOptions.toMap,
+ orCreate = true) // Create the table if it doesn't exist
+
+ case (other, _) =>
+ // We have a potential race condition here in AppendMode, if the table
suddenly gets
+ // created between our existence check and physical execution, but
this can't be helped
+ // in any case.
Review comment:
If we treat `saveAsTable` as one atomic operation, I think AppendMode means
"create or append table". Since "create or append table" is not a standard SQL
commannd, maybe it's fine to ignore this race condition.
----------------------------------------------------------------
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]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
