[HUDI-1451] Support bulk insert v2 with Spark 3.0.0 (#2328)

Co-authored-by: Wenning Ding <wenningd@amazon.com>

- Added support for bulk insert v2 with datasource v2 api in Spark 3.0.0.

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/DefaultSource.java

@@ -0,0 +1,54 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.DataSourceUtils;
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.internal.BaseDefaultSource;
+import org.apache.hudi.internal.DataSourceInternalWriterHelper;
+
+import org.apache.spark.sql.connector.catalog.Table;
+import org.apache.spark.sql.connector.catalog.TableProvider;
+import org.apache.spark.sql.connector.expressions.Transform;
+import org.apache.spark.sql.types.StructType;
+import org.apache.spark.sql.util.CaseInsensitiveStringMap;
+
+import java.util.Map;
+
+/**
+ * DataSource V2 implementation for managing internal write logic. Only called internally.
+ * This class is only compatible with datasource V2 API in Spark 3.
+ */
+public class DefaultSource extends BaseDefaultSource implements TableProvider {
+
+  @Override
+  public StructType inferSchema(CaseInsensitiveStringMap options) {
+    return StructType.fromDDL(options.get(HoodieWriteConfig.BULKINSERT_INPUT_DATA_SCHEMA_DDL));
+  }
+
+  @Override
+  public Table getTable(StructType schema, Transform[] partitioning, Map<String, String> properties) {
+    String instantTime = properties.get(DataSourceInternalWriterHelper.INSTANT_TIME_OPT_KEY);
+    String path = properties.get("path");
+    String tblName = properties.get(HoodieWriteConfig.TABLE_NAME);
+    HoodieWriteConfig config = DataSourceUtils.createHoodieConfig(null, path, tblName, properties);
+    return new HoodieDataSourceInternalTable(instantTime, config, schema, getSparkSession(),
+        getConfiguration());
+  }
+}

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieBulkInsertDataInternalWriter.java

@@ -0,0 +1,64 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.internal.BulkInsertDataInternalWriterHelper;
+import org.apache.hudi.table.HoodieTable;
+
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.connector.write.DataWriter;
+import org.apache.spark.sql.connector.write.WriterCommitMessage;
+import org.apache.spark.sql.types.StructType;
+
+import java.io.IOException;
+
+/**
+ * Hoodie's Implementation of {@link DataWriter<InternalRow>}. This is used in data source "hudi.spark3.internal" implementation for bulk insert.
+ */
+public class HoodieBulkInsertDataInternalWriter implements DataWriter<InternalRow> {
+
+  private final BulkInsertDataInternalWriterHelper bulkInsertWriterHelper;
+
+  public HoodieBulkInsertDataInternalWriter(HoodieTable hoodieTable, HoodieWriteConfig writeConfig,
+      String instantTime, int taskPartitionId, long taskId, StructType structType) {
+    this.bulkInsertWriterHelper = new BulkInsertDataInternalWriterHelper(hoodieTable,
+        writeConfig, instantTime, taskPartitionId, taskId, 0, structType);
+  }
+
+  @Override
+  public void write(InternalRow record) throws IOException {
+    bulkInsertWriterHelper.write(record);
+  }
+
+  @Override
+  public WriterCommitMessage commit() throws IOException {
+    return new HoodieWriterCommitMessage(bulkInsertWriterHelper.getWriteStatuses());
+  }
+
+  @Override
+  public void abort() {
+    bulkInsertWriterHelper.abort();
+  }
+
+  @Override
+  public void close() throws IOException {
+    bulkInsertWriterHelper.close();
+  }
+}

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieBulkInsertDataInternalWriterFactory.java

@@ -0,0 +1,52 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.table.HoodieTable;
+
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.connector.write.DataWriter;
+import org.apache.spark.sql.connector.write.DataWriterFactory;
+import org.apache.spark.sql.types.StructType;
+
+/**
+ * Factory to assist in instantiating {@link HoodieBulkInsertDataInternalWriter}.
+ */
+public class HoodieBulkInsertDataInternalWriterFactory implements DataWriterFactory {
+
+  private final String instantTime;
+  private final HoodieTable hoodieTable;
+  private final HoodieWriteConfig writeConfig;
+  private final StructType structType;
+
+  public HoodieBulkInsertDataInternalWriterFactory(HoodieTable hoodieTable, HoodieWriteConfig writeConfig,
+      String instantTime, StructType structType) {
+    this.hoodieTable = hoodieTable;
+    this.writeConfig = writeConfig;
+    this.instantTime = instantTime;
+    this.structType = structType;
+  }
+
+  @Override
+  public DataWriter<InternalRow> createWriter(int partitionId, long taskId) {
+    return new HoodieBulkInsertDataInternalWriter(hoodieTable, writeConfig, instantTime, partitionId, taskId,
+        structType);
+  }
+}

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieDataSourceInternalBatchWrite.java

@@ -0,0 +1,91 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.client.HoodieInternalWriteStatus;
+import org.apache.hudi.common.model.HoodieWriteStat;
+import org.apache.hudi.common.model.WriteOperationType;
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.internal.DataSourceInternalWriterHelper;
+
+import org.apache.hadoop.conf.Configuration;
+import org.apache.spark.sql.SparkSession;
+import org.apache.spark.sql.connector.write.DataWriterFactory;
+import org.apache.spark.sql.connector.write.WriterCommitMessage;
+import org.apache.spark.sql.connector.write.BatchWrite;
+import org.apache.spark.sql.connector.write.PhysicalWriteInfo;
+import org.apache.spark.sql.types.StructType;
+
+import java.util.Arrays;
+import java.util.List;
+import java.util.stream.Collectors;
+
+/**
+ * Implementation of {@link BatchWrite} for datasource "hudi.spark3.internal" to be used in datasource implementation
+ * of bulk insert.
+ */
+public class HoodieDataSourceInternalBatchWrite implements BatchWrite {
+
+  private final String instantTime;
+  private final HoodieWriteConfig writeConfig;
+  private final StructType structType;
+  private final DataSourceInternalWriterHelper dataSourceInternalWriterHelper;
+
+  public HoodieDataSourceInternalBatchWrite(String instantTime, HoodieWriteConfig writeConfig, StructType structType,
+      SparkSession jss, Configuration hadoopConfiguration) {
+    this.instantTime = instantTime;
+    this.writeConfig = writeConfig;
+    this.structType = structType;
+    this.dataSourceInternalWriterHelper = new DataSourceInternalWriterHelper(instantTime, writeConfig, structType,
+        jss, hadoopConfiguration);
+  }
+
+  @Override
+  public DataWriterFactory createBatchWriterFactory(PhysicalWriteInfo info) {
+    dataSourceInternalWriterHelper.createInflightCommit();
+    if (WriteOperationType.BULK_INSERT == dataSourceInternalWriterHelper.getWriteOperationType()) {
+      return new HoodieBulkInsertDataInternalWriterFactory(dataSourceInternalWriterHelper.getHoodieTable(),
+          writeConfig, instantTime, structType);
+    } else {
+      throw new IllegalArgumentException("Write Operation Type + " + dataSourceInternalWriterHelper.getWriteOperationType() + " not supported ");
+    }
+  }
+
+  @Override
+  public boolean useCommitCoordinator() {
+    return dataSourceInternalWriterHelper.useCommitCoordinator();
+  }
+
+  @Override
+  public void onDataWriterCommit(WriterCommitMessage message) {
+    dataSourceInternalWriterHelper.onDataWriterCommit(message.toString());
+  }
+
+  @Override
+  public void commit(WriterCommitMessage[] messages) {
+    List<HoodieWriteStat> writeStatList = Arrays.stream(messages).map(m -> (HoodieWriterCommitMessage) m)
+        .flatMap(m -> m.getWriteStatuses().stream().map(HoodieInternalWriteStatus::getStat)).collect(Collectors.toList());
+    dataSourceInternalWriterHelper.commit(writeStatList);
+  }
+
+  @Override
+  public void abort(WriterCommitMessage[] messages) {
+    dataSourceInternalWriterHelper.abort();
+  }
+}

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieDataSourceInternalBatchWriteBuilder.java

@@ -0,0 +1,55 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.config.HoodieWriteConfig;
+
+import org.apache.hadoop.conf.Configuration;
+import org.apache.spark.sql.SparkSession;
+import org.apache.spark.sql.connector.write.BatchWrite;
+import org.apache.spark.sql.connector.write.WriteBuilder;
+import org.apache.spark.sql.types.StructType;
+
+/**
+ * Implementation of {@link WriteBuilder} for datasource "hudi.spark3.internal" to be used in datasource implementation
+ * of bulk insert.
+ */
+public class HoodieDataSourceInternalBatchWriteBuilder implements WriteBuilder {
+
+  private final String instantTime;
+  private final HoodieWriteConfig writeConfig;
+  private final StructType structType;
+  private final SparkSession jss;
+  private final Configuration hadoopConfiguration;
+
+  public HoodieDataSourceInternalBatchWriteBuilder(String instantTime, HoodieWriteConfig writeConfig, StructType structType,
+      SparkSession jss, Configuration hadoopConfiguration) {
+    this.instantTime = instantTime;
+    this.writeConfig = writeConfig;
+    this.structType = structType;
+    this.jss = jss;
+    this.hadoopConfiguration = hadoopConfiguration;
+  }
+
+  @Override
+  public BatchWrite buildForBatch() {
+    return new HoodieDataSourceInternalBatchWrite(instantTime, writeConfig, structType, jss,
+        hadoopConfiguration);
+  }
+}

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieDataSourceInternalTable.java

@@ -0,0 +1,77 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hadoop.conf.Configuration;
+import org.apache.hudi.config.HoodieWriteConfig;
+
+import org.apache.spark.sql.SparkSession;
+import org.apache.spark.sql.connector.catalog.SupportsWrite;
+import org.apache.spark.sql.connector.catalog.TableCapability;
+import org.apache.spark.sql.connector.write.LogicalWriteInfo;
+import org.apache.spark.sql.connector.write.WriteBuilder;
+import org.apache.spark.sql.types.StructType;
+
+import java.util.HashSet;
+import java.util.Set;
+
+/**
+ * Hoodie's Implementation of {@link SupportsWrite}. This is used in data source "hudi.spark3.internal" implementation for bulk insert.
+ */
+class HoodieDataSourceInternalTable implements SupportsWrite {
+
+  private final String instantTime;
+  private final HoodieWriteConfig writeConfig;
+  private final StructType structType;
+  private final SparkSession jss;
+  private final Configuration hadoopConfiguration;
+
+  public HoodieDataSourceInternalTable(String instantTime, HoodieWriteConfig config,
+      StructType schema, SparkSession jss, Configuration hadoopConfiguration) {
+    this.instantTime = instantTime;
+    this.writeConfig = config;
+    this.structType = schema;
+    this.jss = jss;
+    this.hadoopConfiguration = hadoopConfiguration;
+  }
+
+  @Override
+  public String name() {
+    return this.getClass().toString();
+  }
+
+  @Override
+  public StructType schema() {
+    return structType;
+  }
+
+  @Override
+  public Set<TableCapability> capabilities() {
+    return new HashSet<TableCapability>() {{
+        add(TableCapability.BATCH_WRITE);
+        add(TableCapability.TRUNCATE);
+      }};
+  }
+
+  @Override
+  public WriteBuilder newWriteBuilder(LogicalWriteInfo logicalWriteInfo) {
+    return new HoodieDataSourceInternalBatchWriteBuilder(instantTime, writeConfig, structType, jss,
+        hadoopConfiguration);
+  }
+}

hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieWriterCommitMessage.java

@@ -0,0 +1,36 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.client.HoodieInternalWriteStatus;
+import org.apache.hudi.internal.BaseWriterCommitMessage;
+import org.apache.spark.sql.connector.write.WriterCommitMessage;
+
+import java.util.List;
+
+/**
+ * Hoodie's {@link WriterCommitMessage} used in datasource "hudi.spark3.internal" implementation.
+ */
+public class HoodieWriterCommitMessage extends BaseWriterCommitMessage
+    implements WriterCommitMessage {
+
+  public HoodieWriterCommitMessage(List<HoodieInternalWriteStatus> writeStatuses) {
+    super(writeStatuses);
+  }
+}

hudi-spark-datasource/hudi-spark3/src/test/scala/org/apache/hudi/spark3/internal/TestHoodieBulkInsertDataInternalWriter.java

@@ -0,0 +1,150 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.common.testutils.HoodieTestDataGenerator;
+import org.apache.hudi.common.util.Option;
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.internal.HoodieBulkInsertInternalWriterTestBase;
+import org.apache.hudi.table.HoodieSparkTable;
+import org.apache.hudi.table.HoodieTable;
+
+import org.apache.spark.sql.Dataset;
+import org.apache.spark.sql.Row;
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.junit.jupiter.api.Test;
+
+import java.util.ArrayList;
+import java.util.List;
+
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.ENCODER;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.STRUCT_TYPE;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.getConfigBuilder;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.getInternalRowWithError;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.getRandomRows;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.toInternalRows;
+import static org.junit.jupiter.api.Assertions.fail;
+
+/**
+ * Unit tests {@link HoodieBulkInsertDataInternalWriter}.
+ */
+public class TestHoodieBulkInsertDataInternalWriter extends
+    HoodieBulkInsertInternalWriterTestBase {
+
+  @Test
+  public void testDataInternalWriter() throws Exception {
+    // init config and table
+    HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
+    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
+    // execute N rounds
+    for (int i = 0; i < 5; i++) {
+      String instantTime = "00" + i;
+      // init writer
+      HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), STRUCT_TYPE);
+
+      int size = 10 + RANDOM.nextInt(1000);
+      // write N rows to partition1, N rows to partition2 and N rows to partition3 ... Each batch should create a new RowCreateHandle and a new file
+      int batches = 5;
+      Dataset<Row> totalInputRows = null;
+
+      for (int j = 0; j < batches; j++) {
+        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
+        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
+        writeRows(inputRows, writer);
+        if (totalInputRows == null) {
+          totalInputRows = inputRows;
+        } else {
+          totalInputRows = totalInputRows.union(inputRows);
+        }
+      }
+
+      HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+      Option<List<String>> fileAbsPaths = Option.of(new ArrayList<>());
+      Option<List<String>> fileNames = Option.of(new ArrayList<>());
+
+      // verify write statuses
+      assertWriteStatuses(commitMetadata.getWriteStatuses(), batches, size, fileAbsPaths, fileNames);
+
+      // verify rows
+      Dataset<Row> result = sqlContext.read().parquet(fileAbsPaths.get().toArray(new String[0]));
+      assertOutput(totalInputRows, result, instantTime, fileNames);
+    }
+  }
+
+
+  /**
+   * Issue some corrupted or wrong schematized InternalRow after few valid InternalRows so that global error is thrown. write batch 1 of valid records write batch2 of invalid records which is expected
+   * to throw Global Error. Verify global error is set appropriately and only first batch of records are written to disk.
+   */
+  @Test
+  public void testGlobalFailure() throws Exception {
+    // init config and table
+    HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
+    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
+    String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[0];
+
+    String instantTime = "001";
+    HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), STRUCT_TYPE);
+
+    int size = 10 + RANDOM.nextInt(100);
+    int totalFailures = 5;
+    // Generate first batch of valid rows
+    Dataset<Row> inputRows = getRandomRows(sqlContext, size / 2, partitionPath, false);
+    List<InternalRow> internalRows = toInternalRows(inputRows, ENCODER);
+
+    // generate some failures rows
+    for (int i = 0; i < totalFailures; i++) {
+      internalRows.add(getInternalRowWithError(partitionPath));
+    }
+
+    // generate 2nd batch of valid rows
+    Dataset<Row> inputRows2 = getRandomRows(sqlContext, size / 2, partitionPath, false);
+    internalRows.addAll(toInternalRows(inputRows2, ENCODER));
+
+    // issue writes
+    try {
+      for (InternalRow internalRow : internalRows) {
+        writer.write(internalRow);
+      }
+      fail("Should have failed");
+    } catch (Throwable e) {
+      // expected
+    }
+
+    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+
+    Option<List<String>> fileAbsPaths = Option.of(new ArrayList<>());
+    Option<List<String>> fileNames = Option.of(new ArrayList<>());
+    // verify write statuses
+    assertWriteStatuses(commitMetadata.getWriteStatuses(), 1, size / 2, fileAbsPaths, fileNames);
+
+    // verify rows
+    Dataset<Row> result = sqlContext.read().parquet(fileAbsPaths.get().toArray(new String[0]));
+    assertOutput(inputRows, result, instantTime, fileNames);
+  }
+
+  private void writeRows(Dataset<Row> inputRows, HoodieBulkInsertDataInternalWriter writer)
+      throws Exception {
+    List<InternalRow> internalRows = toInternalRows(inputRows, ENCODER);
+    // issue writes
+    for (InternalRow internalRow : internalRows) {
+      writer.write(internalRow);
+    }
+  }
+}

hudi-spark-datasource/hudi-spark3/src/test/scala/org/apache/hudi/spark3/internal/TestHoodieDataSourceInternalBatchWrite.java

@@ -0,0 +1,265 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.hudi.spark3.internal;
+
+import org.apache.hudi.common.testutils.HoodieTestDataGenerator;
+import org.apache.hudi.common.util.Option;
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.internal.HoodieBulkInsertInternalWriterTestBase;
+import org.apache.hudi.table.HoodieSparkTable;
+import org.apache.hudi.table.HoodieTable;
+import org.apache.hudi.testutils.HoodieClientTestUtils;
+
+import org.apache.spark.sql.Dataset;
+import org.apache.spark.sql.Row;
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.connector.write.DataWriter;
+import org.junit.jupiter.api.Test;
+
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.List;
+
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.ENCODER;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.STRUCT_TYPE;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.getConfigBuilder;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.getRandomRows;
+import static org.apache.hudi.testutils.SparkDatasetTestUtils.toInternalRows;
+
+/**
+ * Unit tests {@link HoodieDataSourceInternalBatchWrite}.
+ */
+public class TestHoodieDataSourceInternalBatchWrite extends
+    HoodieBulkInsertInternalWriterTestBase {
+
+  @Test
+  public void testDataSourceWriter() throws Exception {
+    // init config and table
+    HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
+    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
+    String instantTime = "001";
+    // init writer
+    HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
+        new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
+    DataWriter<InternalRow> writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(0, RANDOM.nextLong());
+
+    String[] partitionPaths = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS;
+    List<String> partitionPathsAbs = new ArrayList<>();
+    for (String partitionPath : partitionPaths) {
+      partitionPathsAbs.add(basePath + "/" + partitionPath + "/*");
+    }
+
+    int size = 10 + RANDOM.nextInt(1000);
+    int batches = 5;
+    Dataset<Row> totalInputRows = null;
+
+    for (int j = 0; j < batches; j++) {
+      String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
+      Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
+      writeRows(inputRows, writer);
+      if (totalInputRows == null) {
+        totalInputRows = inputRows;
+      } else {
+        totalInputRows = totalInputRows.union(inputRows);
+      }
+    }
+
+    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+    List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
+    commitMessages.add(commitMetadata);
+    dataSourceInternalBatchWrite.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
+    metaClient.reloadActiveTimeline();
+    Dataset<Row> result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
+    // verify output
+    assertOutput(totalInputRows, result, instantTime, Option.empty());
+    assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
+  }
+
+  @Test
+  public void testMultipleDataSourceWrites() throws Exception {
+    // init config and table
+    HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
+    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
+    int partitionCounter = 0;
+
+    // execute N rounds
+    for (int i = 0; i < 5; i++) {
+      String instantTime = "00" + i;
+      // init writer
+      HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
+          new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
+
+      List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
+      Dataset<Row> totalInputRows = null;
+      DataWriter<InternalRow> writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(partitionCounter++, RANDOM.nextLong());
+
+      int size = 10 + RANDOM.nextInt(1000);
+      int batches = 5; // one batch per partition
+
+      for (int j = 0; j < batches; j++) {
+        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
+        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
+        writeRows(inputRows, writer);
+        if (totalInputRows == null) {
+          totalInputRows = inputRows;
+        } else {
+          totalInputRows = totalInputRows.union(inputRows);
+        }
+      }
+
+      HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+      commitMessages.add(commitMetadata);
+      dataSourceInternalBatchWrite.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
+      metaClient.reloadActiveTimeline();
+
+      Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime);
+
+      // verify output
+      assertOutput(totalInputRows, result, instantTime, Option.empty());
+      assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
+    }
+  }
+
+  @Test
+  public void testLargeWrites() throws Exception {
+    // init config and table
+    HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
+    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
+    int partitionCounter = 0;
+
+    // execute N rounds
+    for (int i = 0; i < 3; i++) {
+      String instantTime = "00" + i;
+      // init writer
+      HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
+          new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
+
+      List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
+      Dataset<Row> totalInputRows = null;
+      DataWriter<InternalRow> writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(partitionCounter++, RANDOM.nextLong());
+
+      int size = 10000 + RANDOM.nextInt(10000);
+      int batches = 3; // one batch per partition
+
+      for (int j = 0; j < batches; j++) {
+        String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
+        Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
+        writeRows(inputRows, writer);
+        if (totalInputRows == null) {
+          totalInputRows = inputRows;
+        } else {
+          totalInputRows = totalInputRows.union(inputRows);
+        }
+      }
+
+      HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+      commitMessages.add(commitMetadata);
+      dataSourceInternalBatchWrite.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
+      metaClient.reloadActiveTimeline();
+
+      Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime);
+
+      // verify output
+      assertOutput(totalInputRows, result, instantTime, Option.empty());
+      assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
+    }
+  }
+
+  /**
+   * Tests that DataSourceWriter.abort() will abort the written records of interest write and commit batch1 write and abort batch2 Read of entire dataset should show only records from batch1.
+   * commit batch1
+   * abort batch2
+   * verify only records from batch1 is available to read
+   */
+  @Test
+  public void testAbort() throws Exception {
+    // init config and table
+    HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
+    HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
+    String instantTime0 = "00" + 0;
+    // init writer
+    HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
+        new HoodieDataSourceInternalBatchWrite(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
+
+    DataWriter<InternalRow> writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(0, RANDOM.nextLong());
+
+    List<String> partitionPaths = Arrays.asList(HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS);
+    List<String> partitionPathsAbs = new ArrayList<>();
+    for (String partitionPath : partitionPaths) {
+      partitionPathsAbs.add(basePath + "/" + partitionPath + "/*");
+    }
+
+    int size = 10 + RANDOM.nextInt(100);
+    int batches = 1;
+    Dataset<Row> totalInputRows = null;
+
+    for (int j = 0; j < batches; j++) {
+      String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
+      Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
+      writeRows(inputRows, writer);
+      if (totalInputRows == null) {
+        totalInputRows = inputRows;
+      } else {
+        totalInputRows = totalInputRows.union(inputRows);
+      }
+    }
+
+    HoodieWriterCommitMessage commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+    List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
+    commitMessages.add(commitMetadata);
+    // commit 1st batch
+    dataSourceInternalBatchWrite.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
+    metaClient.reloadActiveTimeline();
+    Dataset<Row> result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
+    // verify rows
+    assertOutput(totalInputRows, result, instantTime0, Option.empty());
+    assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
+
+    // 2nd batch. abort in the end
+    String instantTime1 = "00" + 1;
+    dataSourceInternalBatchWrite =
+        new HoodieDataSourceInternalBatchWrite(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
+    writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(1, RANDOM.nextLong());
+
+    for (int j = 0; j < batches; j++) {
+      String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[j % 3];
+      Dataset<Row> inputRows = getRandomRows(sqlContext, size, partitionPath, false);
+      writeRows(inputRows, writer);
+    }
+
+    commitMetadata = (HoodieWriterCommitMessage) writer.commit();
+    commitMessages = new ArrayList<>();
+    commitMessages.add(commitMetadata);
+    // commit 1st batch
+    dataSourceInternalBatchWrite.abort(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
+    metaClient.reloadActiveTimeline();
+    result = HoodieClientTestUtils.read(jsc, basePath, sqlContext, metaClient.getFs(), partitionPathsAbs.toArray(new String[0]));
+    // verify rows
+    // only rows from first batch should be present
+    assertOutput(totalInputRows, result, instantTime0, Option.empty());
+  }
+
+  private void writeRows(Dataset<Row> inputRows, DataWriter<InternalRow> writer) throws Exception {
+    List<InternalRow> internalRows = toInternalRows(inputRows, ENCODER);
+    // issue writes
+    for (InternalRow internalRow : internalRows) {
+      writer.write(internalRow);
+    }
+  }
+}