yihua commented on code in PR #13882:
URL: https://github.com/apache/hudi/pull/13882#discussion_r2345680041
##########
hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/io/storage/row/HoodieRowParquetWriteSupport.java:
##########
@@ -18,47 +18,149 @@
package org.apache.hudi.io.storage.row;
+import org.apache.hudi.HoodieSparkUtils;
+import org.apache.hudi.SparkAdapterSupport;
+import org.apache.hudi.avro.HoodieAvroUtils;
import org.apache.hudi.avro.HoodieBloomFilterWriteSupport;
import org.apache.hudi.common.bloom.BloomFilter;
import org.apache.hudi.common.config.HoodieConfig;
import org.apache.hudi.common.config.HoodieStorageConfig;
import org.apache.hudi.common.util.Option;
import org.apache.hudi.common.util.ReflectionUtils;
+import org.apache.hudi.config.HoodieWriteConfig;
+import org.apache.hudi.internal.schema.convert.AvroInternalSchemaConverter;
+import org.apache.hudi.internal.schema.utils.SerDeHelper;
+import org.apache.avro.LogicalType;
+import org.apache.avro.LogicalTypes;
+import org.apache.avro.Schema;
import org.apache.hadoop.conf.Configuration;
+import org.apache.parquet.avro.AvroSchemaConverter;
+import org.apache.parquet.avro.AvroWriteSupport;
import org.apache.parquet.hadoop.api.WriteSupport;
+import org.apache.parquet.io.api.Binary;
+import org.apache.parquet.io.api.RecordConsumer;
+import org.apache.parquet.schema.MessageType;
+import org.apache.spark.sql.HoodieInternalRowUtils;
+import org.apache.spark.sql.catalyst.InternalRow;
+import org.apache.spark.sql.catalyst.expressions.SpecializedGetters;
+import org.apache.spark.sql.catalyst.util.ArrayData;
+import org.apache.spark.sql.catalyst.util.DateTimeUtils;
+import org.apache.spark.sql.catalyst.util.MapData;
+import org.apache.spark.sql.execution.datasources.DataSourceUtils;
import org.apache.spark.sql.execution.datasources.parquet.ParquetWriteSupport;
+import org.apache.spark.sql.internal.SQLConf;
+import org.apache.spark.sql.types.ArrayType;
+import org.apache.spark.sql.types.DataType;
+import org.apache.spark.sql.types.DataTypes;
+import org.apache.spark.sql.types.Decimal;
+import org.apache.spark.sql.types.DecimalType;
+import org.apache.spark.sql.types.MapType;
import org.apache.spark.sql.types.StructType;
+import org.apache.spark.sql.types.TimestampNTZType;
import org.apache.spark.unsafe.types.UTF8String;
+import org.apache.spark.util.VersionUtils;
import java.util.Collections;
+import java.util.HashMap;
import java.util.Map;
+import java.util.stream.IntStream;
+import scala.Enumeration;
+import scala.Function1;
+
+import static org.apache.hudi.avro.AvroSchemaUtils.resolveNullableSchema;
import static
org.apache.hudi.common.config.HoodieStorageConfig.PARQUET_FIELD_ID_WRITE_ENABLED;
/**
- * Hoodie Write Support for directly writing Row to Parquet.
+ * Hoodie Write Support for directly writing Row to Parquet and adding the
Hudi bloom index to the file metadata.
+ * The implementation is adapted from Spark's {@link
org.apache.spark.sql.execution.datasources.parquet.ParquetWriteSupport} but has
the following major differences:
+ * <ul>
+ * <li>Decimals are always written with the legacy format to ensure
compatibility with parquet-avro and other readers</li>
+ * <li>Writing 2-level or 3-level lists is configurable with either the Hudi
option, hoodie.parquet.writelegacyformat.enabled, or the parquet-avro option,
parquet.avro.write-old-list-structure
+ * to ensure consistency across writer paths.</li>
+ * <li>The scale of the timestamps is determined by the Hudi writer schema
instead of relying on Spark configuration</li>
+ * </ul>
*/
-public class HoodieRowParquetWriteSupport extends ParquetWriteSupport {
+public class HoodieRowParquetWriteSupport extends WriteSupport<InternalRow> {
+ private static final Schema MAP_KEY_SCHEMA =
Schema.create(Schema.Type.STRING);
+ private static final String MAP_REPEATED_NAME = "key_value";
+ private static final String MAP_KEY_NAME = "key";
+ private static final String MAP_VALUE_NAME = "value";
private final Configuration hadoopConf;
private final Option<HoodieBloomFilterWriteSupport<UTF8String>>
bloomFilterWriteSupportOpt;
+ private final byte[] decimalBuffer = new
byte[Decimal.minBytesForPrecision()[DecimalType.MAX_PRECISION()]];
+ private final Enumeration.Value datetimeRebaseMode = (Enumeration.Value)
SparkAdapterSupport.getSparkAdapter().getDateTimeRebaseMode();
+ private final Function1<Object, Object> dateRebaseFunction =
DataSourceUtils.createDateRebaseFuncInWrite(datetimeRebaseMode, "Parquet");
+ private final Function1<Object, Object> timestampRebaseFunction =
DataSourceUtils.createTimestampRebaseFuncInWrite(datetimeRebaseMode, "Parquet");
+ private RecordConsumer recordConsumer;
+ private final boolean writeLegacyListFormat;
+ private final ValueWriter[] rootFieldWriters;
+ private final Schema avroSchema;
- public HoodieRowParquetWriteSupport(Configuration conf, StructType
structType, Option<BloomFilter> bloomFilterOpt, HoodieConfig config) {
+ public HoodieRowParquetWriteSupport(Configuration conf, StructType schema,
Option<BloomFilter> bloomFilterOpt, HoodieWriteConfig config) {
Configuration hadoopConf = new Configuration(conf);
- hadoopConf.set("spark.sql.parquet.writeLegacyFormat",
config.getStringOrDefault(HoodieStorageConfig.PARQUET_WRITE_LEGACY_FORMAT_ENABLED,
"false"));
+ String writeLegacyFormatEnabled =
config.getStringOrDefault(HoodieStorageConfig.PARQUET_WRITE_LEGACY_FORMAT_ENABLED,
"false");
+ hadoopConf.set("spark.sql.parquet.writeLegacyFormat",
writeLegacyFormatEnabled);
hadoopConf.set("spark.sql.parquet.outputTimestampType",
config.getStringOrDefault(HoodieStorageConfig.PARQUET_OUTPUT_TIMESTAMP_TYPE));
hadoopConf.set("spark.sql.parquet.fieldId.write.enabled",
config.getStringOrDefault(PARQUET_FIELD_ID_WRITE_ENABLED));
- setSchema(structType, hadoopConf);
-
+ this.writeLegacyListFormat = Boolean.parseBoolean(writeLegacyFormatEnabled)
+ ||
Boolean.parseBoolean(config.getStringOrDefault(AvroWriteSupport.WRITE_OLD_LIST_STRUCTURE,
"false"));
+ this.avroSchema =
SerDeHelper.fromJson(config.getInternalSchema()).map(internalSchema ->
AvroInternalSchemaConverter.convert(internalSchema, "spark_schema"))
+ .orElseGet(() -> {
+ Schema parsedSchema = new
Schema.Parser().parse(config.getWriteSchema());
+ if (config.populateMetaFields()) {
+ return HoodieAvroUtils.addMetadataFields(parsedSchema,
config.allowOperationMetadataField());
+ }
+ return parsedSchema;
+ });
+
ParquetWriteSupport.setSchema(HoodieInternalRowUtils.getCachedSchema(avroSchema),
hadoopConf);
+ this.rootFieldWriters = getFieldWriters(schema, avroSchema);
this.hadoopConf = hadoopConf;
this.bloomFilterWriteSupportOpt =
bloomFilterOpt.map(HoodieBloomFilterRowWriteSupport::new);
}
+ private ValueWriter[] getFieldWriters(StructType schema, Schema avroSchema) {
+ Map<String, Integer> fieldNameToIndex = new HashMap<>();
+ for (int i = 0; i < schema.fields().length; i++) {
+ fieldNameToIndex.put(schema.fields()[i].name(), i);
+ }
+ return IntStream.range(0, avroSchema.getFields().size()).mapToObj(i -> {
+ Schema.Field field = avroSchema.getFields().get(i);
+ Integer structIndex = fieldNameToIndex.get(field.name());
+ return makeWriter(field.schema(),
schema.fields()[structIndex].dataType());
+ }).toArray(ValueWriter[]::new);
+ }
+
public Configuration getHadoopConf() {
return hadoopConf;
}
+ @Override
+ public WriteContext init(Configuration configuration) {
+ Map<String, String> metadata = new HashMap<>();
+ metadata.put("org.apache.spark.version",
VersionUtils.shortVersion(HoodieSparkUtils.getSparkVersion()));
Review Comment:
`org.apache.hudi.parquet.write.support` to have both version and class name?
@vinothchandar do you think this is OK?
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
