lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

[HUDI-1104] Adding support for UserDefinedPartitioners and SortModes to BulkInsert with Rows (#3149)

Browse Source
This commit is contained in:
Sivabalan Narayanan
2021-07-07 11:15:25 -04:00
committed by GitHub
parent 55ecbc662e
commit ea9e5d0e8b
31 changed files with 618 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/DefaultSource.java
View File

@@ -19,6 +19,7 @@
package org.apache.hudi.spark3.internal;
import org.apache.hudi.DataSourceUtils;
import org.apache.hudi.config.HoodieInternalConfig;
import org.apache.hudi.config.HoodieWriteConfig;
import org.apache.hudi.internal.BaseDefaultSource;
import org.apache.hudi.internal.DataSourceInternalWriterHelper;
@@ -47,9 +48,11 @@ public class DefaultSource extends BaseDefaultSource implements TableProvider {
String instantTime = properties.get(DataSourceInternalWriterHelper.INSTANT_TIME_OPT_KEY);
String path = properties.get("path");
String tblName = properties.get(HoodieWriteConfig.TABLE_NAME.key());
boolean arePartitionRecordsSorted = Boolean.parseBoolean(properties.getOrDefault(HoodieInternalConfig.BULKINSERT_ARE_PARTITIONER_RECORDS_SORTED,
Boolean.toString(HoodieInternalConfig.DEFAULT_BULKINSERT_ARE_PARTITIONER_RECORDS_SORTED)));
// 1st arg to createHooodieConfig is not really reuqired to be set. but passing it anyways.
HoodieWriteConfig config = DataSourceUtils.createHoodieConfig(properties.get(HoodieWriteConfig.AVRO_SCHEMA.key()), path, tblName, properties);
return new HoodieDataSourceInternalTable(instantTime, config, schema, getSparkSession(),
getConfiguration());
getConfiguration(), arePartitionRecordsSorted);
}
}

4
hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieBulkInsertDataInternalWriter.java
View File

@@ -37,9 +37,9 @@ public class HoodieBulkInsertDataInternalWriter implements DataWriter<InternalRo
private final BulkInsertDataInternalWriterHelper bulkInsertWriterHelper;
public HoodieBulkInsertDataInternalWriter(HoodieTable hoodieTable, HoodieWriteConfig writeConfig,
String instantTime, int taskPartitionId, long taskId, StructType structType) {
String instantTime, int taskPartitionId, long taskId, StructType structType, boolean arePartitionRecordsSorted) {
this.bulkInsertWriterHelper = new BulkInsertDataInternalWriterHelper(hoodieTable,
writeConfig, instantTime, taskPartitionId, taskId, 0, structType);
writeConfig, instantTime, taskPartitionId, taskId, 0, structType, arePartitionRecordsSorted);
}
@Override

6
hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieBulkInsertDataInternalWriterFactory.java
View File

@@ -35,18 +35,20 @@ public class HoodieBulkInsertDataInternalWriterFactory implements DataWriterFact
private final HoodieTable hoodieTable;
private final HoodieWriteConfig writeConfig;
private final StructType structType;
private final boolean arePartitionRecordsSorted;
public HoodieBulkInsertDataInternalWriterFactory(HoodieTable hoodieTable, HoodieWriteConfig writeConfig,
String instantTime, StructType structType) {
String instantTime, StructType structType, boolean arePartitionRecordsSorted) {
this.hoodieTable = hoodieTable;
this.writeConfig = writeConfig;
this.instantTime = instantTime;
this.structType = structType;
this.arePartitionRecordsSorted = arePartitionRecordsSorted;
}
@Override
public DataWriter<InternalRow> createWriter(int partitionId, long taskId) {
return new HoodieBulkInsertDataInternalWriter(hoodieTable, writeConfig, instantTime, partitionId, taskId,
structType);
structType, arePartitionRecordsSorted);
}
}

6
hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieDataSourceInternalBatchWrite.java
View File

@@ -45,13 +45,15 @@ public class HoodieDataSourceInternalBatchWrite implements BatchWrite {
private final String instantTime;
private final HoodieWriteConfig writeConfig;
private final StructType structType;
private final boolean arePartitionRecordsSorted;
private final DataSourceInternalWriterHelper dataSourceInternalWriterHelper;
public HoodieDataSourceInternalBatchWrite(String instantTime, HoodieWriteConfig writeConfig, StructType structType,
SparkSession jss, Configuration hadoopConfiguration) {
SparkSession jss, Configuration hadoopConfiguration, boolean arePartitionRecordsSorted) {
this.instantTime = instantTime;
this.writeConfig = writeConfig;
this.structType = structType;
this.arePartitionRecordsSorted = arePartitionRecordsSorted;
this.dataSourceInternalWriterHelper = new DataSourceInternalWriterHelper(instantTime, writeConfig, structType,
jss, hadoopConfiguration);
}
@@ -61,7 +63,7 @@ public class HoodieDataSourceInternalBatchWrite implements BatchWrite {
dataSourceInternalWriterHelper.createInflightCommit();
if (WriteOperationType.BULK_INSERT == dataSourceInternalWriterHelper.getWriteOperationType()) {
return new HoodieBulkInsertDataInternalWriterFactory(dataSourceInternalWriterHelper.getHoodieTable(),
writeConfig, instantTime, structType);
writeConfig, instantTime, structType, arePartitionRecordsSorted);
} else {
throw new IllegalArgumentException("Write Operation Type + " + dataSourceInternalWriterHelper.getWriteOperationType() + " not supported ");
}

6
hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieDataSourceInternalBatchWriteBuilder.java
View File

@@ -37,19 +37,21 @@ public class HoodieDataSourceInternalBatchWriteBuilder implements WriteBuilder {
private final StructType structType;
private final SparkSession jss;
private final Configuration hadoopConfiguration;
private final boolean arePartitionRecordsSorted;
public HoodieDataSourceInternalBatchWriteBuilder(String instantTime, HoodieWriteConfig writeConfig, StructType structType,
SparkSession jss, Configuration hadoopConfiguration) {
SparkSession jss, Configuration hadoopConfiguration, boolean arePartitionRecordsSorted) {
this.instantTime = instantTime;
this.writeConfig = writeConfig;
this.structType = structType;
this.jss = jss;
this.hadoopConfiguration = hadoopConfiguration;
this.arePartitionRecordsSorted = arePartitionRecordsSorted;
}
@Override
public BatchWrite buildForBatch() {
return new HoodieDataSourceInternalBatchWrite(instantTime, writeConfig, structType, jss,
hadoopConfiguration);
hadoopConfiguration, arePartitionRecordsSorted);
}
}

6
hudi-spark-datasource/hudi-spark3/src/main/java/org/apache/hudi/spark3/internal/HoodieDataSourceInternalTable.java
View File

@@ -41,14 +41,16 @@ class HoodieDataSourceInternalTable implements SupportsWrite {
private final StructType structType;
private final SparkSession jss;
private final Configuration hadoopConfiguration;
private final boolean arePartitionRecordsSorted;
public HoodieDataSourceInternalTable(String instantTime, HoodieWriteConfig config,
StructType schema, SparkSession jss, Configuration hadoopConfiguration) {
StructType schema, SparkSession jss, Configuration hadoopConfiguration, boolean arePartitionRecordsSorted) {
this.instantTime = instantTime;
this.writeConfig = config;
this.structType = schema;
this.jss = jss;
this.hadoopConfiguration = hadoopConfiguration;
this.arePartitionRecordsSorted = arePartitionRecordsSorted;
}
@Override
@@ -73,6 +75,6 @@ class HoodieDataSourceInternalTable implements SupportsWrite {
@Override
public WriteBuilder newWriteBuilder(LogicalWriteInfo logicalWriteInfo) {
return new HoodieDataSourceInternalBatchWriteBuilder(instantTime, writeConfig, structType, jss,
hadoopConfiguration);
hadoopConfiguration, arePartitionRecordsSorted);
}
}

24
hudi-spark-datasource/hudi-spark3/src/test/java/org/apache/hudi/spark3/internal/TestHoodieBulkInsertDataInternalWriter.java
View File

@@ -29,9 +29,13 @@ 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 org.junit.jupiter.params.ParameterizedTest;
import org.junit.jupiter.params.provider.Arguments;
import org.junit.jupiter.params.provider.MethodSource;
import java.util.ArrayList;
import java.util.List;
import java.util.stream.Stream;
import static org.apache.hudi.testutils.SparkDatasetTestUtils.ENCODER;
import static org.apache.hudi.testutils.SparkDatasetTestUtils.STRUCT_TYPE;
@@ -47,8 +51,17 @@ import static org.junit.jupiter.api.Assertions.fail;
public class TestHoodieBulkInsertDataInternalWriter extends
HoodieBulkInsertInternalWriterTestBase {
@Test
public void testDataInternalWriter() throws Exception {
private static Stream<Arguments> configParams() {
Object[][] data = new Object[][] {
{true},
{false}
};
return Stream.of(data).map(Arguments::of);
}
@ParameterizedTest
@MethodSource("configParams")
public void testDataInternalWriter(boolean sorted) throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
@@ -56,7 +69,8 @@ public class TestHoodieBulkInsertDataInternalWriter extends
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);
HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), STRUCT_TYPE,
sorted);
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
@@ -79,7 +93,7 @@ public class TestHoodieBulkInsertDataInternalWriter extends
Option<List<String>> fileNames = Option.of(new ArrayList<>());
// verify write statuses
assertWriteStatuses(commitMetadata.getWriteStatuses(), batches, size, fileAbsPaths, fileNames);
assertWriteStatuses(commitMetadata.getWriteStatuses(), batches, size, sorted, fileAbsPaths, fileNames);
// verify rows
Dataset<Row> result = sqlContext.read().parquet(fileAbsPaths.get().toArray(new String[0]));
@@ -100,7 +114,7 @@ public class TestHoodieBulkInsertDataInternalWriter extends
String partitionPath = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS[0];
String instantTime = "001";
HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), STRUCT_TYPE);
HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), STRUCT_TYPE, false);
int size = 10 + RANDOM.nextInt(100);
int totalFailures = 5;

10
hudi-spark-datasource/hudi-spark3/src/test/java/org/apache/hudi/spark3/internal/TestHoodieDataSourceInternalBatchWrite.java
View File

@@ -56,7 +56,7 @@ public class TestHoodieDataSourceInternalBatchWrite extends
String instantTime = "001";
// init writer
HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, false);
DataWriter<InternalRow> writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(0, RANDOM.nextLong());
String[] partitionPaths = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS;
@@ -103,7 +103,7 @@ public class TestHoodieDataSourceInternalBatchWrite extends
String instantTime = "00" + i;
// init writer
HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, false);
List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
Dataset<Row> totalInputRows = null;
@@ -148,7 +148,7 @@ public class TestHoodieDataSourceInternalBatchWrite extends
String instantTime = "00" + i;
// init writer
HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
new HoodieDataSourceInternalBatchWrite(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, false);
List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
Dataset<Row> totalInputRows = null;
@@ -195,7 +195,7 @@ public class TestHoodieDataSourceInternalBatchWrite extends
String instantTime0 = "00" + 0;
// init writer
HoodieDataSourceInternalBatchWrite dataSourceInternalBatchWrite =
new HoodieDataSourceInternalBatchWrite(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
new HoodieDataSourceInternalBatchWrite(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, false);
DataWriter<InternalRow> writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(0, RANDOM.nextLong());
@@ -234,7 +234,7 @@ public class TestHoodieDataSourceInternalBatchWrite extends
// 2nd batch. abort in the end
String instantTime1 = "00" + 1;
dataSourceInternalBatchWrite =
new HoodieDataSourceInternalBatchWrite(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf);
new HoodieDataSourceInternalBatchWrite(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, false);
writer = dataSourceInternalBatchWrite.createBatchWriterFactory(null).createWriter(1, RANDOM.nextLong());
for (int j = 0; j < batches; j++) {