Github user cloud-fan commented on a diff in the pull request: https://github.com/apache/spark/pull/15996#discussion_r93722231 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/DataFrameWriter.scala --- @@ -364,48 +366,162 @@ final class DataFrameWriter[T] private[sql](ds: Dataset[T]) { throw new AnalysisException("Cannot create hive serde table with saveAsTable API") } - val tableExists = df.sparkSession.sessionState.catalog.tableExists(tableIdent) - - (tableExists, mode) match { - case (true, SaveMode.Ignore) => - // Do nothing - - case (true, SaveMode.ErrorIfExists) => - throw new AnalysisException(s"Table $tableIdent already exists.") - - case _ => - val existingTable = if (tableExists) { - Some(df.sparkSession.sessionState.catalog.getTableMetadata(tableIdent)) - } else { - None - } - val storage = if (tableExists) { - existingTable.get.storage - } else { - DataSource.buildStorageFormatFromOptions(extraOptions.toMap) - } - val tableType = if (tableExists) { - existingTable.get.tableType - } else if (storage.locationUri.isDefined) { - CatalogTableType.EXTERNAL - } else { - CatalogTableType.MANAGED + val catalog = df.sparkSession.sessionState.catalog + val db = tableIdent.database.getOrElse(catalog.getCurrentDatabase) + val tableIdentWithDB = tableIdent.copy(database = Some(db)) + val tableName = tableIdentWithDB.unquotedString + + catalog.getTableMetadataOption(tableIdentWithDB) match { + // If the table already exists... + case Some(existingTable) => + mode match { + case SaveMode.Ignore => // Do nothing + + case SaveMode.ErrorIfExists => + throw new AnalysisException(s"Table $tableName already exists. You can set SaveMode " + + "to SaveMode.Append to insert data into the table or set SaveMode to " + + "SaveMode.Overwrite to overwrite the existing data.") + + case SaveMode.Append => + if (existingTable.tableType == CatalogTableType.VIEW) { + throw new AnalysisException("Saving data into a view is not allowed.") + } + + if (existingTable.provider.get == DDLUtils.HIVE_PROVIDER) { + throw new AnalysisException(s"Saving data in the Hive serde table $tableName is " + + "not supported yet. Please use the insertInto() API as an alternative.") + } + + // Check if the specified data source match the data source of the existing table. + val existingProvider = DataSource.lookupDataSource(existingTable.provider.get) + val specifiedProvider = DataSource.lookupDataSource(source) + // TODO: Check that options from the resolved relation match the relation that we are + // inserting into (i.e. using the same compression). + if (existingProvider != specifiedProvider) { + throw new AnalysisException(s"The format of the existing table $tableName is " + + s"`${existingProvider.getSimpleName}`. It doesn't match the specified format " + + s"`${specifiedProvider.getSimpleName}`.") + } + + if (df.schema.length != existingTable.schema.length) { + throw new AnalysisException( + s"The column number of the existing table $tableName" + + s"(${existingTable.schema.catalogString}) doesn't match the data schema" + + s"(${df.schema.catalogString})") + } + + val resolver = df.sparkSession.sessionState.conf.resolver + val tableCols = existingTable.schema.map(_.name) + + // As we are inserting into an existing table, we should respect the existing schema and + // adjust the column order of the given dataframe according to it, or throw exception + // if the column names do not match. + val adjustedColumns = tableCols.map { col => + df.queryExecution.analyzed.resolve(Seq(col), resolver).getOrElse { + val inputColumns = df.schema.map(_.name).mkString(", ") + throw new AnalysisException( + s"cannot resolve '$col' given input columns: [$inputColumns]") + } + } + + // Check if the specified partition columns match the existing table. + val specifiedPartCols = CatalogUtils.normalizePartCols( + tableName, tableCols, partitioningColumns.getOrElse(Nil), resolver) + if (specifiedPartCols != existingTable.partitionColumnNames) { + val existingPartCols = existingTable.partitionColumnNames.mkString(", ") + throw new AnalysisException( + s""" + |Specified partitioning does not match that of the existing table $tableName. + |Specified partition columns: [${specifiedPartCols.mkString(", ")}] + |Existing partition columns: [$existingPartCols] + """.stripMargin) + } + + // Check if the specified bucketing match the existing table. + val specifiedBucketSpec = getBucketSpec.map { bucketSpec => + CatalogUtils.normalizeBucketSpec(tableName, tableCols, bucketSpec, resolver) + } + if (specifiedBucketSpec != existingTable.bucketSpec) { + val specifiedBucketString = + specifiedBucketSpec.map(_.toString).getOrElse("not bucketed") + val existingBucketString = + existingTable.bucketSpec.map(_.toString).getOrElse("not bucketed") + throw new AnalysisException( + s""" + |Specified bucketing does not match that of the existing table $tableName. + |Specified bucketing: $specifiedBucketString + |Existing bucketing: $existingBucketString + """.stripMargin) + } + + // Reorder the columns of the given dataframe to match the existing table. + val adjustedDataFrame = df.select(adjustedColumns.map(new Column(_)): _*) + + if (classOf[CreatableRelationProvider].isAssignableFrom(existingProvider)) { + // For data source that implementing `CreatableRelationProvider`, we should call its + // `createRelation` to do the data appending. `CreatableRelationProvider` may not + // provide an `InsertableRelation`, so we can't use `DataFrameWriter.insertInto` here. + existingProvider.newInstance().asInstanceOf[CreatableRelationProvider].createRelation( + sqlContext = df.sparkSession.sqlContext, + mode = mode, + parameters = new CaseInsensitiveMap(extraOptions.toMap), + data = adjustedDataFrame) + } else { + adjustedDataFrame.write.insertInto(tableIdentWithDB) --- End diff -- Currently only 2 data sources accept `saveAsTable` with append mode: `CreatableRelationProvider` and `FileFormat`. For `CreatableRelationProvider`, we always go to `createRelation`, for `FileFormat`, we go to `InsertIntoHadoopFsRelation`, which is same with `InsertIntoTable`. That's why I add the if-else here.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org