[SPARK-26356][SQL] remove SaveMode from data source v2 #24233
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -30,12 +30,11 @@ import org.apache.spark.sql.catalyst.plans.logical.{AppendData, InsertIntoTable, | |
| import org.apache.spark.sql.execution.SQLExecution | ||
| import org.apache.spark.sql.execution.command.DDLUtils | ||
| import org.apache.spark.sql.execution.datasources.{CreateTable, DataSource, DataSourceUtils, LogicalRelation} | ||
| import org.apache.spark.sql.execution.datasources.v2._ | ||
| import org.apache.spark.sql.internal.SQLConf | ||
| import org.apache.spark.sql.sources.BaseRelation | ||
| import org.apache.spark.sql.sources.v2._ | ||
| import org.apache.spark.sql.sources.v2.TableCapability._ | ||
| import org.apache.spark.sql.types.StructType | ||
| import org.apache.spark.sql.util.CaseInsensitiveStringMap | ||
|
|
||
|
|
@@ -56,13 +55,16 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| * <li>`SaveMode.Overwrite`: overwrite the existing data.</li> | ||
| * <li>`SaveMode.Append`: append the data.</li> | ||
| * <li>`SaveMode.Ignore`: ignore the operation (i.e. no-op).</li> | ||
| * <li>`SaveMode.ErrorIfExists`: throw an exception at runtime.</li> | ||
| * </ul> | ||
| * <p> | ||
| * When writing to data source v1, the default option is `ErrorIfExists`. When writing to data | ||
| * source v2, the default option is `Append`. | ||
| * | ||
| * @since 1.4.0 | ||
| */ | ||
| def mode(saveMode: SaveMode): DataFrameWriter[T] = { | ||
| this.mode = Some(saveMode) | ||
| this | ||
| } | ||
|
|
||
|
|
@@ -78,15 +80,15 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| * @since 1.4.0 | ||
| */ | ||
| def mode(saveMode: String): DataFrameWriter[T] = { | ||
| saveMode.toLowerCase(Locale.ROOT) match { | ||
| case "overwrite" => mode(SaveMode.Overwrite) | ||
| case "append" => mode(SaveMode.Append) | ||
| case "ignore" => mode(SaveMode.Ignore) | ||
| case "error" | "errorifexists" => mode(SaveMode.ErrorIfExists) | ||
| case "default" => this | ||
|
|
||
| case _ => throw new IllegalArgumentException(s"Unknown save mode: $saveMode. " + | ||
| "Accepted save modes are 'overwrite', 'append', 'ignore', 'error', 'errorifexists'.") | ||
| } | ||
| } | ||
|
|
||
| /** | ||
|
|
@@ -268,9 +270,24 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
|
|
||
| import org.apache.spark.sql.execution.datasources.v2.DataSourceV2Implicits._ | ||
| provider.getTable(dsOptions) match { | ||
| // TODO (SPARK-27815): To not break existing tests, here we treat file source as a special | ||
| // case, and pass the save mode to file source directly. This hack should be removed. | ||
| case table: FileTable => | ||
|
There was a problem hiding this comment. Why is this hack necessary? Why not put off v2 support for path-based tables? There was a problem hiding this comment. Because we want to migrate file source to v2, to validate the API of This hack is just to work around the issue that we do not have a proper entry API for path-based data source, which I believe we will have later on. I think this is not a bad idea. We can unblock the file source migration, and we can keep the DS v2 clean ( There was a problem hiding this comment. File sources do not use v2 by default, so this is not a necessary change for this commit. I think it should be in a separate PR. We can discuss whether it is a good idea to add this hack in the next DSv2 sync. Please remove it from this PR. There was a problem hiding this comment. I'm adding this to my agenda for the sync-up. There was a problem hiding this comment. I believe that we agreed in the sync-up to remove this hack. Is there a reason why it is still included? There was a problem hiding this comment. I explained in the comment: #24233 (comment) Removing the hack breaks several tests and I'd like to do it in another PR. Since |
||
| val write = table.newWriteBuilder(dsOptions).asInstanceOf[FileWriteBuilder] | ||
| .mode(modeForDSV1) // should not change default mode for file source. | ||
| .withQueryId(UUID.randomUUID().toString) | ||
| .withInputDataSchema(df.logicalPlan.schema) | ||
| .buildForBatch() | ||
| // The returned `Write` can be null, which indicates that we can skip writing. | ||
| if (write != null) { | ||
| runCommand(df.sparkSession, "save") { | ||
| WriteToDataSourceV2(write, df.logicalPlan) | ||
| } | ||
| } | ||
|
|
||
| case table: SupportsWrite if table.supports(BATCH_WRITE) => | ||
| lazy val relation = DataSourceV2Relation.create(table, dsOptions) | ||
| modeForDSV2 match { | ||
| case SaveMode.Append => | ||
| runCommand(df.sparkSession, "save") { | ||
| AppendData.byName(relation, df.logicalPlan) | ||
|
|
@@ -282,25 +299,10 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| OverwriteByExpression.byName(relation, df.logicalPlan, Literal(true)) | ||
| } | ||
|
|
||
| case other => | ||
| throw new AnalysisException(s"TableProvider implementation $source cannot be " + | ||
| s"written with $other mode, please use Append or Overwrite " + | ||
| "modes instead.") | ||
| } | ||
|
|
||
| // Streaming also uses the data source V2 API. So it may be that the data source implements | ||
|
|
@@ -328,7 +330,7 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| sparkSession = df.sparkSession, | ||
| className = source, | ||
| partitionColumns = partitioningColumns.getOrElse(Nil), | ||
| options = extraOptions.toMap).planForWriting(modeForDSV1, df.logicalPlan) | ||
| } | ||
| } | ||
|
|
||
|
|
@@ -377,7 +379,7 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| table = UnresolvedRelation(tableIdent), | ||
| partition = Map.empty[String, Option[String]], | ||
| query = df.logicalPlan, | ||
| overwrite = modeForDSV1 == SaveMode.Overwrite, | ||
| ifPartitionNotExists = false) | ||
| } | ||
| } | ||
|
|
@@ -457,7 +459,7 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| val tableIdentWithDB = tableIdent.copy(database = Some(db)) | ||
| val tableName = tableIdentWithDB.unquotedString | ||
|
|
||
| (tableExists, modeForDSV1) match { | ||
| case (true, SaveMode.Ignore) => | ||
| // Do nothing | ||
|
|
||
|
|
@@ -512,7 +514,8 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| partitionColumnNames = partitioningColumns.getOrElse(Nil), | ||
| bucketSpec = getBucketSpec) | ||
|
|
||
| runCommand(df.sparkSession, "saveAsTable")( | ||
| CreateTable(tableDesc, modeForDSV1, Some(df.logicalPlan))) | ||
| } | ||
|
|
||
| /** | ||
|
|
@@ -718,13 +721,17 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { | |
| SQLExecution.withNewExecutionId(session, qe, Some(name))(qe.toRdd) | ||
| } | ||
|
|
||
| private def modeForDSV1 = mode.getOrElse(SaveMode.ErrorIfExists) | ||
|
|
||
| private def modeForDSV2 = mode.getOrElse(SaveMode.Append) | ||
|
|
||
| /////////////////////////////////////////////////////////////////////////////////////// | ||
| // Builder pattern config options | ||
| /////////////////////////////////////////////////////////////////////////////////////// | ||
|
|
||
| private var source: String = df.sparkSession.sessionState.conf.defaultDataSourceName | ||
|
|
||
| private var mode: Option[SaveMode] = None | ||
|
|
||
| private val extraOptions = new scala.collection.mutable.HashMap[String, String] | ||
|
|
||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think @rxin objected to changing the default mode for DSv2. Has that changed?
I think that this is a good idea because the most sensible default for v2 is to append if no mode is set, instead of failing because the mode is not supported. This doesn't change v1 behavior, so I think it is okay.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I think @rxin has a concern about breaking existing ETL pipelines. Now we only change the default mode for DS v2(exclude file sources), I think there is no compatibility issue.