Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[HUDI-7006] Reduce unnecessary is_empty rdd calls in StreamSync #10158

Merged
merged 3 commits into from
Nov 25, 2023
Merged

[HUDI-7006] Reduce unnecessary is_empty rdd calls in StreamSync #10158

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
08794fc
Reduce unnecessary is_empty rdd calls in StreamSync
harsh1231 Nov 22, 2023
40237b5
Fixing test failures
nsivabalan Nov 23, 2023
d81ae35
Fix testParamsWithSchemaTransformer to check nested exceptions for me…
harsh1231 Nov 24, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion ...-integ-test/src/main/java/org/apache/hudi/integ/testsuite/HoodieDeltaStreamerWrapper.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,7 +87,7 @@ public Pair<SchemaProvider, Pair<String, JavaRDD<HoodieRecord>>> fetchSource() t
.setBasePath(service.getCfg().targetBasePath)
.build();
String instantTime = InProcessTimeGenerator.createNewInstantTime();
InputBatch inputBatch = service.readFromSource(instantTime, metaClient).getLeft();
InputBatch inputBatch = service.readFromSource(instantTime, metaClient);
return Pair.of(inputBatch.getSchemaProvider(), Pair.of(inputBatch.getCheckpointForNextBatch(), (JavaRDD<HoodieRecord>) inputBatch.getBatch().get()));
}

Expand Down
100 changes: 50 additions & 50 deletions hudi-utilities/src/main/java/org/apache/hudi/utilities/streamer/HoodieStreamerUtils.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -70,63 +70,63 @@ public class HoodieStreamerUtils {
* Takes care of dropping columns, precombine, auto key generation.
* Both AVRO and SPARK record types are supported.
*/
static JavaRDD<HoodieRecord> createHoodieRecords(HoodieStreamer.Config cfg, TypedProperties props, Option<JavaRDD<GenericRecord>> avroRDDOptional,
static Option<JavaRDD<HoodieRecord>> createHoodieRecords(HoodieStreamer.Config cfg, TypedProperties props, Option<JavaRDD<GenericRecord>> avroRDDOptional,
SchemaProvider schemaProvider, HoodieRecord.HoodieRecordType recordType, boolean autoGenerateRecordKeys,
String instantTime) {
boolean shouldCombine = cfg.filterDupes || cfg.operation.equals(WriteOperationType.UPSERT);
Set<String> partitionColumns = getPartitionColumns(props);
JavaRDD<GenericRecord> avroRDD = avroRDDOptional.get();
return avroRDDOptional.map(avroRDD -> {
JavaRDD<HoodieRecord> records;
SerializableSchema avroSchema = new SerializableSchema(schemaProvider.getTargetSchema());
SerializableSchema processedAvroSchema = new SerializableSchema(isDropPartitionColumns(props) ? HoodieAvroUtils.removeMetadataFields(avroSchema.get()) : avroSchema.get());
if (recordType == HoodieRecord.HoodieRecordType.AVRO) {
records = avroRDD.mapPartitions(
(FlatMapFunction<Iterator<GenericRecord>, HoodieRecord>) genericRecordIterator -> {
if (autoGenerateRecordKeys) {
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_PARTITION_ID_CONFIG, String.valueOf(TaskContext.getPartitionId()));
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_INSTANT_TIME_CONFIG, instantTime);
}
BuiltinKeyGenerator builtinKeyGenerator = (BuiltinKeyGenerator) HoodieSparkKeyGeneratorFactory.createKeyGenerator(props);
List<HoodieRecord> avroRecords = new ArrayList<>();
while (genericRecordIterator.hasNext()) {
GenericRecord genRec = genericRecordIterator.next();
HoodieKey hoodieKey = new HoodieKey(builtinKeyGenerator.getRecordKey(genRec), builtinKeyGenerator.getPartitionPath(genRec));
GenericRecord gr = isDropPartitionColumns(props) ? HoodieAvroUtils.removeFields(genRec, partitionColumns) : genRec;
HoodieRecordPayload payload = shouldCombine ? DataSourceUtils.createPayload(cfg.payloadClassName, gr,
(Comparable) HoodieAvroUtils.getNestedFieldVal(gr, cfg.sourceOrderingField, false, props.getBoolean(
KeyGeneratorOptions.KEYGENERATOR_CONSISTENT_LOGICAL_TIMESTAMP_ENABLED.key(),
Boolean.parseBoolean(KeyGeneratorOptions.KEYGENERATOR_CONSISTENT_LOGICAL_TIMESTAMP_ENABLED.defaultValue()))))
: DataSourceUtils.createPayload(cfg.payloadClassName, gr);
avroRecords.add(new HoodieAvroRecord<>(hoodieKey, payload));
}
return avroRecords.iterator();
});
} else if (recordType == HoodieRecord.HoodieRecordType.SPARK) {
// TODO we should remove it if we can read InternalRow from source.
records = avroRDD.mapPartitions(itr -> {
if (autoGenerateRecordKeys) {
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_PARTITION_ID_CONFIG, String.valueOf(TaskContext.getPartitionId()));
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_INSTANT_TIME_CONFIG, instantTime);
}
BuiltinKeyGenerator builtinKeyGenerator = (BuiltinKeyGenerator) HoodieSparkKeyGeneratorFactory.createKeyGenerator(props);
StructType baseStructType = AvroConversionUtils.convertAvroSchemaToStructType(processedAvroSchema.get());
StructType targetStructType = isDropPartitionColumns(props) ? AvroConversionUtils
.convertAvroSchemaToStructType(HoodieAvroUtils.removeFields(processedAvroSchema.get(), partitionColumns)) : baseStructType;
HoodieAvroDeserializer deserializer = SparkAdapterSupport$.MODULE$.sparkAdapter().createAvroDeserializer(processedAvroSchema.get(), baseStructType);

JavaRDD<HoodieRecord> records;
SerializableSchema avroSchema = new SerializableSchema(schemaProvider.getTargetSchema());
SerializableSchema processedAvroSchema = new SerializableSchema(isDropPartitionColumns(props) ? HoodieAvroUtils.removeMetadataFields(avroSchema.get()) : avroSchema.get());
if (recordType == HoodieRecord.HoodieRecordType.AVRO) {
records = avroRDD.mapPartitions(
(FlatMapFunction<Iterator<GenericRecord>, HoodieRecord>) genericRecordIterator -> {
if (autoGenerateRecordKeys) {
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_PARTITION_ID_CONFIG, String.valueOf(TaskContext.getPartitionId()));
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_INSTANT_TIME_CONFIG, instantTime);
}
BuiltinKeyGenerator builtinKeyGenerator = (BuiltinKeyGenerator) HoodieSparkKeyGeneratorFactory.createKeyGenerator(props);
List<HoodieRecord> avroRecords = new ArrayList<>();
while (genericRecordIterator.hasNext()) {
GenericRecord genRec = genericRecordIterator.next();
HoodieKey hoodieKey = new HoodieKey(builtinKeyGenerator.getRecordKey(genRec), builtinKeyGenerator.getPartitionPath(genRec));
GenericRecord gr = isDropPartitionColumns(props) ? HoodieAvroUtils.removeFields(genRec, partitionColumns) : genRec;
HoodieRecordPayload payload = shouldCombine ? DataSourceUtils.createPayload(cfg.payloadClassName, gr,
(Comparable) HoodieAvroUtils.getNestedFieldVal(gr, cfg.sourceOrderingField, false, props.getBoolean(
KeyGeneratorOptions.KEYGENERATOR_CONSISTENT_LOGICAL_TIMESTAMP_ENABLED.key(),
Boolean.parseBoolean(KeyGeneratorOptions.KEYGENERATOR_CONSISTENT_LOGICAL_TIMESTAMP_ENABLED.defaultValue()))))
: DataSourceUtils.createPayload(cfg.payloadClassName, gr);
avroRecords.add(new HoodieAvroRecord<>(hoodieKey, payload));
}
return avroRecords.iterator();
return new CloseableMappingIterator<>(ClosableIterator.wrap(itr), rec -> {
InternalRow row = (InternalRow) deserializer.deserialize(rec).get();
String recordKey = builtinKeyGenerator.getRecordKey(row, baseStructType).toString();
String partitionPath = builtinKeyGenerator.getPartitionPath(row, baseStructType).toString();
return new HoodieSparkRecord(new HoodieKey(recordKey, partitionPath),
HoodieInternalRowUtils.getCachedUnsafeProjection(baseStructType, targetStructType).apply(row), targetStructType, false);
});
} else if (recordType == HoodieRecord.HoodieRecordType.SPARK) {
// TODO we should remove it if we can read InternalRow from source.
records = avroRDD.mapPartitions(itr -> {
if (autoGenerateRecordKeys) {
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_PARTITION_ID_CONFIG, String.valueOf(TaskContext.getPartitionId()));
props.setProperty(KeyGenUtils.RECORD_KEY_GEN_INSTANT_TIME_CONFIG, instantTime);
}
BuiltinKeyGenerator builtinKeyGenerator = (BuiltinKeyGenerator) HoodieSparkKeyGeneratorFactory.createKeyGenerator(props);
StructType baseStructType = AvroConversionUtils.convertAvroSchemaToStructType(processedAvroSchema.get());
StructType targetStructType = isDropPartitionColumns(props) ? AvroConversionUtils
.convertAvroSchemaToStructType(HoodieAvroUtils.removeFields(processedAvroSchema.get(), partitionColumns)) : baseStructType;
HoodieAvroDeserializer deserializer = SparkAdapterSupport$.MODULE$.sparkAdapter().createAvroDeserializer(processedAvroSchema.get(), baseStructType);

return new CloseableMappingIterator<>(ClosableIterator.wrap(itr), rec -> {
InternalRow row = (InternalRow) deserializer.deserialize(rec).get();
String recordKey = builtinKeyGenerator.getRecordKey(row, baseStructType).toString();
String partitionPath = builtinKeyGenerator.getPartitionPath(row, baseStructType).toString();
return new HoodieSparkRecord(new HoodieKey(recordKey, partitionPath),
HoodieInternalRowUtils.getCachedUnsafeProjection(baseStructType, targetStructType).apply(row), targetStructType, false);
});
});
} else {
throw new UnsupportedOperationException(recordType.name());
}
return records;
} else {
throw new UnsupportedOperationException(recordType.name());
}
return records;
});
}

/**
Expand Down
44 changes: 25 additions & 19 deletions hudi-utilities/src/main/java/org/apache/hudi/utilities/streamer/SparkSampleWritesUtils.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,7 +64,7 @@ public class SparkSampleWritesUtils {

private static final Logger LOG = LoggerFactory.getLogger(SparkSampleWritesUtils.class);

public static Option<HoodieWriteConfig> getWriteConfigWithRecordSizeEstimate(JavaSparkContext jsc, JavaRDD<HoodieRecord> records, HoodieWriteConfig writeConfig) {
public static Option<HoodieWriteConfig> getWriteConfigWithRecordSizeEstimate(JavaSparkContext jsc, Option<JavaRDD<HoodieRecord>> recordsOpt, HoodieWriteConfig writeConfig) {
if (!writeConfig.getBoolean(SAMPLE_WRITES_ENABLED)) {
LOG.debug("Skip overwriting record size estimate as it's disabled.");
return Option.empty();
Expand All @@ -76,7 +76,7 @@ public static Option<HoodieWriteConfig> getWriteConfigWithRecordSizeEstimate(Jav
}
try {
String instantTime = getInstantFromTemporalAccessor(Instant.now().atZone(ZoneId.systemDefault()));
Pair<Boolean, String> result = doSampleWrites(jsc, records, writeConfig, instantTime);
Pair<Boolean, String> result = doSampleWrites(jsc, recordsOpt, writeConfig, instantTime);
if (result.getLeft()) {
long avgSize = getAvgSizeFromSampleWrites(jsc, result.getRight());
LOG.info("Overwriting record size estimate to " + avgSize);
Expand All @@ -90,7 +90,7 @@ public static Option<HoodieWriteConfig> getWriteConfigWithRecordSizeEstimate(Jav
return Option.empty();
}

private static Pair<Boolean, String> doSampleWrites(JavaSparkContext jsc, JavaRDD<HoodieRecord> records, HoodieWriteConfig writeConfig, String instantTime)
private static Pair<Boolean, String> doSampleWrites(JavaSparkContext jsc, Option<JavaRDD<HoodieRecord>> recordsOpt, HoodieWriteConfig writeConfig, String instantTime)
throws IOException {
final String sampleWritesBasePath = getSampleWritesBasePath(jsc, writeConfig, instantTime);
HoodieTableMetaClient.withPropertyBuilder()
Expand All @@ -109,25 +109,31 @@ private static Pair<Boolean, String> doSampleWrites(JavaSparkContext jsc, JavaRD
.withAutoCommit(true)
.withPath(sampleWritesBasePath)
.build();
Pair<Boolean, String> emptyRes = Pair.of(false, null);
try (SparkRDDWriteClient sampleWriteClient = new SparkRDDWriteClient(new HoodieSparkEngineContext(jsc), sampleWriteConfig, Option.empty())) {
int size = writeConfig.getIntOrDefault(SAMPLE_WRITES_SIZE);
List<HoodieRecord> samples = records.coalesce(1).take(size);
sampleWriteClient.startCommitWithTime(instantTime);
JavaRDD<WriteStatus> writeStatusRDD = sampleWriteClient.bulkInsert(jsc.parallelize(samples, 1), instantTime);
if (writeStatusRDD.filter(WriteStatus::hasErrors).count() > 0) {
LOG.error(String.format("sample writes for table %s failed with errors.", writeConfig.getTableName()));
if (LOG.isTraceEnabled()) {
LOG.trace("Printing out the top 100 errors");
writeStatusRDD.filter(WriteStatus::hasErrors).take(100).forEach(ws -> {
LOG.trace("Global error :", ws.getGlobalError());
ws.getErrors().forEach((key, throwable) ->
LOG.trace(String.format("Error for key: %s", key), throwable));
});
return recordsOpt.map(records -> {
List<HoodieRecord> samples = records.coalesce(1).take(size);
if (samples.isEmpty()) {
return emptyRes;
}
return Pair.of(false, null);
} else {
return Pair.of(true, sampleWritesBasePath);
}
sampleWriteClient.startCommitWithTime(instantTime);
JavaRDD<WriteStatus> writeStatusRDD = sampleWriteClient.bulkInsert(jsc.parallelize(samples, 1), instantTime);
if (writeStatusRDD.filter(WriteStatus::hasErrors).count() > 0) {
LOG.error(String.format("sample writes for table %s failed with errors.", writeConfig.getTableName()));
if (LOG.isTraceEnabled()) {
LOG.trace("Printing out the top 100 errors");
writeStatusRDD.filter(WriteStatus::hasErrors).take(100).forEach(ws -> {
LOG.trace("Global error :", ws.getGlobalError());
ws.getErrors().forEach((key, throwable) ->
LOG.trace(String.format("Error for key: %s", key), throwable));
});
}
return emptyRes;
} else {
return Pair.of(true, sampleWritesBasePath);
}
}).orElse(emptyRes);
}
}

Expand Down
Loading
Loading