lanyuanxiaoyao/hudi
1
0
Fork 0
Code Activity

[HUDI-2161] Adding support to disable meta columns with bulk insert operation (#3247)

Browse Source
This commit is contained in:
Sivabalan Narayanan
2021-07-19 20:43:48 -04:00
committed by GitHub
parent 2099bf41db
commit d5026e9a24
53 changed files with 1063 additions and 269 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
hudi-spark-datasource/hudi-spark2/src/main/java/org/apache/hudi/internal/DefaultSource.java
View File

@@ -19,6 +19,7 @@
package org.apache.hudi.internal;
import org.apache.hudi.DataSourceUtils;
import org.apache.hudi.common.table.HoodieTableConfig;
import org.apache.hudi.config.HoodieInternalConfig;
import org.apache.hudi.config.HoodieWriteConfig;
@@ -61,12 +62,14 @@ public class DefaultSource extends BaseDefaultSource implements DataSourceV2,
String instantTime = options.get(DataSourceInternalWriterHelper.INSTANT_TIME_OPT_KEY).get();
String path = options.get("path").get();
String tblName = options.get(HoodieWriteConfig.TABLE_NAME.key()).get();
boolean populateMetaFields = options.getBoolean(HoodieTableConfig.HOODIE_POPULATE_META_FIELDS.key(),
Boolean.parseBoolean(HoodieTableConfig.HOODIE_POPULATE_META_FIELDS.defaultValue()));
// 1st arg to createHooodieConfig is not really reuqired to be set. but passing it anyways.
HoodieWriteConfig config = DataSourceUtils.createHoodieConfig(options.get(HoodieWriteConfig.AVRO_SCHEMA.key()).get(), path, tblName, options.asMap());
boolean arePartitionRecordsSorted = HoodieInternalConfig.getBulkInsertIsPartitionRecordsSorted(
options.get(HoodieInternalConfig.BULKINSERT_ARE_PARTITIONER_RECORDS_SORTED).isPresent()
? options.get(HoodieInternalConfig.BULKINSERT_ARE_PARTITIONER_RECORDS_SORTED).get() : null);
return Optional.of(new HoodieDataSourceInternalWriter(instantTime, config, schema, getSparkSession(),
getConfiguration(), options, arePartitionRecordsSorted));
getConfiguration(), options, populateMetaFields, arePartitionRecordsSorted));
}
}

6
hudi-spark-datasource/hudi-spark2/src/main/java/org/apache/hudi/internal/HoodieBulkInsertDataInternalWriter.java
View File

@@ -36,10 +36,10 @@ public class HoodieBulkInsertDataInternalWriter implements DataWriter<InternalRo
private final BulkInsertDataInternalWriterHelper bulkInsertWriterHelper;
public HoodieBulkInsertDataInternalWriter(HoodieTable hoodieTable, HoodieWriteConfig writeConfig,
String instantTime, int taskPartitionId, long taskId, long taskEpochId,
StructType structType, boolean arePartitionRecordsSorted) {
String instantTime, int taskPartitionId, long taskId, long taskEpochId,
StructType structType, boolean populateMetaFields, boolean arePartitionRecordsSorted) {
this.bulkInsertWriterHelper = new BulkInsertDataInternalWriterHelper(hoodieTable,
writeConfig, instantTime, taskPartitionId, taskId, taskEpochId, structType, arePartitionRecordsSorted);
writeConfig, instantTime, taskPartitionId, taskId, taskEpochId, structType, populateMetaFields, arePartitionRecordsSorted);
}
@Override

7
hudi-spark-datasource/hudi-spark2/src/main/java/org/apache/hudi/internal/HoodieBulkInsertDataInternalWriterFactory.java
View File

@@ -36,19 +36,22 @@ public class HoodieBulkInsertDataInternalWriterFactory implements DataWriterFact
private final HoodieWriteConfig writeConfig;
private final StructType structType;
private final boolean arePartitionRecordsSorted;
private final boolean populateMetaFields;
public HoodieBulkInsertDataInternalWriterFactory(HoodieTable hoodieTable, HoodieWriteConfig writeConfig,
String instantTime, StructType structType, boolean arePartitionRecordsSorted) {
String instantTime, StructType structType, boolean populateMetaFields,
boolean arePartitionRecordsSorted) {
this.hoodieTable = hoodieTable;
this.writeConfig = writeConfig;
this.instantTime = instantTime;
this.structType = structType;
this.populateMetaFields = populateMetaFields;
this.arePartitionRecordsSorted = arePartitionRecordsSorted;
}
@Override
public DataWriter<InternalRow> createDataWriter(int partitionId, long taskId, long epochId) {
return new HoodieBulkInsertDataInternalWriter(hoodieTable, writeConfig, instantTime, partitionId, taskId, epochId,
structType, arePartitionRecordsSorted);
structType, populateMetaFields, arePartitionRecordsSorted);
}
}

6
hudi-spark-datasource/hudi-spark2/src/main/java/org/apache/hudi/internal/HoodieDataSourceInternalWriter.java
View File

@@ -49,15 +49,17 @@ public class HoodieDataSourceInternalWriter implements DataSourceWriter {
private final HoodieWriteConfig writeConfig;
private final StructType structType;
private final DataSourceInternalWriterHelper dataSourceInternalWriterHelper;
private final boolean populateMetaFields;
private final Boolean arePartitionRecordsSorted;
private Map<String, String> extraMetadataMap = new HashMap<>();
public HoodieDataSourceInternalWriter(String instantTime, HoodieWriteConfig writeConfig, StructType structType,
SparkSession sparkSession, Configuration configuration, DataSourceOptions dataSourceOptions,
boolean arePartitionRecordsSorted) {
boolean populateMetaFields, boolean arePartitionRecordsSorted) {
this.instantTime = instantTime;
this.writeConfig = writeConfig;
this.structType = structType;
this.populateMetaFields = populateMetaFields;
this.arePartitionRecordsSorted = arePartitionRecordsSorted;
this.extraMetadataMap = DataSourceUtils.getExtraMetadata(dataSourceOptions.asMap());
this.dataSourceInternalWriterHelper = new DataSourceInternalWriterHelper(instantTime, writeConfig, structType,
@@ -69,7 +71,7 @@ public class HoodieDataSourceInternalWriter implements DataSourceWriter {
dataSourceInternalWriterHelper.createInflightCommit();
if (WriteOperationType.BULK_INSERT == dataSourceInternalWriterHelper.getWriteOperationType()) {
return new HoodieBulkInsertDataInternalWriterFactory(dataSourceInternalWriterHelper.getHoodieTable(),
writeConfig, instantTime, structType, arePartitionRecordsSorted);
writeConfig, instantTime, structType, populateMetaFields, arePartitionRecordsSorted);
} else {
throw new IllegalArgumentException("Write Operation Type + " + dataSourceInternalWriterHelper.getWriteOperationType() + " not supported ");
}

30
hudi-spark-datasource/hudi-spark2/src/test/java/org/apache/hudi/internal/TestHoodieBulkInsertDataInternalWriter.java
View File

@@ -38,7 +38,6 @@ import java.util.stream.Stream;
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;
@@ -51,6 +50,16 @@ public class TestHoodieBulkInsertDataInternalWriter extends
HoodieBulkInsertInternalWriterTestBase {
private static Stream<Arguments> configParams() {
Object[][] data = new Object[][] {
{true, true},
{true, false},
{false, true},
{false, false}
};
return Stream.of(data).map(Arguments::of);
}
private static Stream<Arguments> bulkInsertTypeParams() {
Object[][] data = new Object[][] {
{true},
{false}
@@ -60,16 +69,16 @@ public class TestHoodieBulkInsertDataInternalWriter extends
@ParameterizedTest
@MethodSource("configParams")
public void testDataInternalWriter(boolean sorted) throws Exception {
public void testDataInternalWriter(boolean sorted, boolean populateMetaFields) throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
HoodieTable table = HoodieSparkTable.create(cfg, context, metaClient);
// execute N rounds
for (int i = 0; i < 3; i++) {
String instantTime = "00" + i;
// init writer
HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), RANDOM.nextLong(), STRUCT_TYPE,
sorted);
HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), RANDOM.nextLong(),
STRUCT_TYPE, populateMetaFields, 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
@@ -96,11 +105,10 @@ public class TestHoodieBulkInsertDataInternalWriter extends
// verify rows
Dataset<Row> result = sqlContext.read().parquet(fileAbsPaths.get().toArray(new String[0]));
assertOutput(totalInputRows, result, instantTime, fileNames);
assertOutput(totalInputRows, result, instantTime, fileNames, populateMetaFields);
}
}
/**
* 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.
@@ -108,13 +116,13 @@ public class TestHoodieBulkInsertDataInternalWriter extends
@Test
public void testGlobalFailure() throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieWriteConfig cfg = getWriteConfig(true);
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(), RANDOM.nextLong(), STRUCT_TYPE,
false);
HoodieBulkInsertDataInternalWriter writer = new HoodieBulkInsertDataInternalWriter(table, cfg, instantTime, RANDOM.nextInt(100000), RANDOM.nextLong(), RANDOM.nextLong(),
STRUCT_TYPE, true, false);
int size = 10 + RANDOM.nextInt(100);
int totalFailures = 5;
@@ -150,7 +158,7 @@ public class TestHoodieBulkInsertDataInternalWriter extends
// verify rows
Dataset<Row> result = sqlContext.read().parquet(fileAbsPaths.get().toArray(new String[0]));
assertOutput(inputRows, result, instantTime, fileNames);
assertOutput(inputRows, result, instantTime, fileNames, true);
}
private void writeRows(Dataset<Row> inputRows, HoodieBulkInsertDataInternalWriter writer)

76
hudi-spark-datasource/hudi-spark2/src/test/java/org/apache/hudi/internal/TestHoodieDataSourceInternalWriter.java
View File

@@ -31,6 +31,9 @@ import org.apache.spark.sql.catalyst.InternalRow;
import org.apache.spark.sql.sources.v2.DataSourceOptions;
import org.apache.spark.sql.sources.v2.writer.DataWriter;
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.Arrays;
@@ -38,10 +41,10 @@ import java.util.Collections;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.stream.Stream;
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;
import static org.junit.jupiter.api.Assertions.assertEquals;
@@ -53,18 +56,28 @@ import static org.junit.jupiter.api.Assertions.assertTrue;
public class TestHoodieDataSourceInternalWriter extends
HoodieBulkInsertInternalWriterTestBase {
@Test
public void testDataSourceWriter() throws Exception {
testDataSourceWriterInternal(Collections.EMPTY_MAP, Collections.EMPTY_MAP);
private static Stream<Arguments> bulkInsertTypeParams() {
Object[][] data = new Object[][] {
{true},
{false}
};
return Stream.of(data).map(Arguments::of);
}
private void testDataSourceWriterInternal(Map<String, String> extraMetadata, Map<String, String> expectedExtraMetadata) throws Exception {
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testDataSourceWriter(boolean populateMetaFields) throws Exception {
testDataSourceWriterInternal(Collections.EMPTY_MAP, Collections.EMPTY_MAP, populateMetaFields);
}
private void testDataSourceWriterInternal(Map<String, String> extraMetadata, Map<String, String> expectedExtraMetadata, boolean populateMetaFields)
throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
String instantTime = "001";
// init writer
HoodieDataSourceInternalWriter dataSourceInternalWriter =
new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(extraMetadata), false);
new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(extraMetadata), populateMetaFields, false);
DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(0, RANDOM.nextLong(), RANDOM.nextLong());
String[] partitionPaths = HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS;
@@ -95,7 +108,7 @@ public class TestHoodieDataSourceInternalWriter extends
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());
assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
// verify extra metadata
@@ -122,7 +135,7 @@ public class TestHoodieDataSourceInternalWriter extends
expectedMetadata.remove(DataSourceWriteOptions.COMMIT_METADATA_KEYPREFIX_OPT_KEY().key());
expectedMetadata.remove("commit_extra_c");
testDataSourceWriterInternal(extraMeta, expectedMetadata);
testDataSourceWriterInternal(extraMeta, expectedMetadata, true);
}
@Test
@@ -134,13 +147,14 @@ public class TestHoodieDataSourceInternalWriter extends
extraMeta.put("keyB", "valB");
extraMeta.put("commit_extra_c", "valC");
// none of the keys has commit metadata key prefix.
testDataSourceWriterInternal(extraMeta, Collections.EMPTY_MAP);
testDataSourceWriterInternal(extraMeta, Collections.EMPTY_MAP, true);
}
@Test
public void testMultipleDataSourceWrites() throws Exception {
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testMultipleDataSourceWrites(boolean populateMetaFields) throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
int partitionCounter = 0;
// execute N rounds
@@ -148,7 +162,7 @@ public class TestHoodieDataSourceInternalWriter extends
String instantTime = "00" + i;
// init writer
HoodieDataSourceInternalWriter dataSourceInternalWriter =
new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), false);
new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
Dataset<Row> totalInputRows = null;
DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(partitionCounter++, RANDOM.nextLong(), RANDOM.nextLong());
@@ -172,18 +186,20 @@ public class TestHoodieDataSourceInternalWriter extends
dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
metaClient.reloadActiveTimeline();
Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime);
Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime,
populateMetaFields);
// verify output
assertOutput(totalInputRows, result, instantTime, Option.empty());
assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
}
}
@Test
public void testLargeWrites() throws Exception {
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testLargeWrites(boolean populateMetaFields) throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
int partitionCounter = 0;
// execute N rounds
@@ -191,7 +207,7 @@ public class TestHoodieDataSourceInternalWriter extends
String instantTime = "00" + i;
// init writer
HoodieDataSourceInternalWriter dataSourceInternalWriter =
new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), false);
new HoodieDataSourceInternalWriter(instantTime, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
List<HoodieWriterCommitMessage> commitMessages = new ArrayList<>();
Dataset<Row> totalInputRows = null;
DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(partitionCounter++, RANDOM.nextLong(), RANDOM.nextLong());
@@ -215,10 +231,11 @@ public class TestHoodieDataSourceInternalWriter extends
dataSourceInternalWriter.commit(commitMessages.toArray(new HoodieWriterCommitMessage[0]));
metaClient.reloadActiveTimeline();
Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime);
Dataset<Row> result = HoodieClientTestUtils.readCommit(basePath, sqlContext, metaClient.getCommitTimeline(), instantTime,
populateMetaFields);
// verify output
assertOutput(totalInputRows, result, instantTime, Option.empty());
assertOutput(totalInputRows, result, instantTime, Option.empty(), populateMetaFields);
assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
}
}
@@ -229,15 +246,16 @@ public class TestHoodieDataSourceInternalWriter extends
* abort batch2
* verify only records from batch1 is available to read
*/
@Test
public void testAbort() throws Exception {
@ParameterizedTest
@MethodSource("bulkInsertTypeParams")
public void testAbort(boolean populateMetaFields) throws Exception {
// init config and table
HoodieWriteConfig cfg = getConfigBuilder(basePath).build();
HoodieWriteConfig cfg = getWriteConfig(populateMetaFields);
String instantTime0 = "00" + 0;
// init writer
HoodieDataSourceInternalWriter dataSourceInternalWriter =
new HoodieDataSourceInternalWriter(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), false);
new HoodieDataSourceInternalWriter(instantTime0, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
DataWriter<InternalRow> writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(0, RANDOM.nextLong(), RANDOM.nextLong());
List<String> partitionPaths = Arrays.asList(HoodieTestDataGenerator.DEFAULT_PARTITION_PATHS);
@@ -269,13 +287,13 @@ public class TestHoodieDataSourceInternalWriter extends
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());
assertOutput(totalInputRows, result, instantTime0, Option.empty(), populateMetaFields);
assertWriteStatuses(commitMessages.get(0).getWriteStatuses(), batches, size, Option.empty(), Option.empty());
// 2nd batch. abort in the end
String instantTime1 = "00" + 1;
dataSourceInternalWriter =
new HoodieDataSourceInternalWriter(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), false);
new HoodieDataSourceInternalWriter(instantTime1, cfg, STRUCT_TYPE, sqlContext.sparkSession(), hadoopConf, new DataSourceOptions(Collections.EMPTY_MAP), populateMetaFields, false);
writer = dataSourceInternalWriter.createWriterFactory().createDataWriter(1, RANDOM.nextLong(), RANDOM.nextLong());
for (int j = 0; j < batches; j++) {
@@ -293,7 +311,7 @@ public class TestHoodieDataSourceInternalWriter extends
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());
assertOutput(totalInputRows, result, instantTime0, Option.empty(), populateMetaFields);
}
private void writeRows(Dataset<Row> inputRows, DataWriter<InternalRow> writer) throws Exception {