Timestamp millis repair

hudi-client/hudi-java-client/src/test/java/org/apache/hudi/hadoop/TestHoodieFileGroupReaderOnHive.java

@@ -27,6 +27,8 @@
 import org.apache.hudi.common.testutils.HoodieTestDataGenerator;
 import org.apache.hudi.common.testutils.HoodieTestUtils;
 import org.apache.hudi.common.testutils.minicluster.HdfsTestService;
+import org.apache.hudi.common.util.Option;
+import org.apache.hudi.hadoop.avro.HoodieTimestampAwareParquetInputFormat;
 import org.apache.hudi.storage.HoodieStorage;
 import org.apache.hudi.storage.StorageConfiguration;
 import org.apache.hudi.storage.hadoop.HoodieHadoopStorage;
@@ -37,6 +39,7 @@
 import org.apache.hadoop.fs.FileSystem;
 import org.apache.hadoop.hive.metastore.api.hive_metastoreConstants;
 import org.apache.hadoop.hive.ql.io.parquet.MapredParquetInputFormat;
+import org.apache.hadoop.hive.ql.io.parquet.read.ParquetRecordReaderWrapper;
 import org.apache.hadoop.hive.serde2.ColumnProjectionUtils;
 import org.apache.hadoop.hive.serde2.SerDeException;
 import org.apache.hadoop.hive.serde2.avro.HiveTypeUtils;
@@ -94,7 +97,18 @@ public StorageConfiguration<?> getStorageConf() {
 
   @Override
   public HoodieReaderContext<ArrayWritable> getHoodieReaderContext(String tablePath, Schema avroSchema, StorageConfiguration<?> storageConf, HoodieTableMetaClient metaClient) {
-    HoodieFileGroupReaderBasedRecordReader.HiveReaderCreator readerCreator = (inputSplit, jobConf) -> new MapredParquetInputFormat().getRecordReader(inputSplit, jobConf, null);
+    HoodieFileGroupReaderBasedRecordReader.HiveReaderCreator readerCreator = (inputSplit, jobConf, dataSchema) -> {
+      if (HoodieColumnProjectionUtils.supportTimestamp(jobConf)) {
+        try {
+          return new ParquetRecordReaderWrapper(new HoodieTimestampAwareParquetInputFormat(Option.empty(), Option.of(dataSchema)), inputSplit, jobConf, null);
+        } catch (InterruptedException e) {
+          throw new RuntimeException(e);
+        }
+      } else {
+        return new MapredParquetInputFormat().getRecordReader(inputSplit, jobConf, null);
+      }
+    };
+
     JobConf jobConf = new JobConf(storageConf.unwrapAs(Configuration.class));
     setupJobconf(jobConf, avroSchema);
     return new HiveHoodieReaderContext(readerCreator,

hudi-client/hudi-java-client/src/test/java/org/apache/hudi/testutils/ArrayWritableTestUtil.java

@@ -23,6 +23,7 @@
 import org.apache.avro.Schema;
 import org.apache.hadoop.hive.serde2.io.DateWritable;
 import org.apache.hadoop.hive.serde2.io.HiveDecimalWritable;
+import org.apache.hadoop.hive.serde2.io.TimestampWritable;
 import org.apache.hadoop.io.ArrayWritable;
 import org.apache.hadoop.io.BooleanWritable;
 import org.apache.hadoop.io.BytesWritable;
@@ -268,7 +269,11 @@ private static void assertWritablePrimaryTypeMatchesSchema(Schema schema, Writab
         break;
 
       case LONG:
-        assertInstanceOf(LongWritable.class, writable);
+        if (schema.getLogicalType() instanceof LogicalTypes.TimestampMillis) {
+          assertInstanceOf(TimestampWritable.class, writable);
+        } else  {
+          assertInstanceOf(LongWritable.class, writable);
+        }
         break;
 
       case FLOAT:

hudi-client/hudi-spark-client/src/main/java/org/apache/hudi/io/storage/HoodieSparkParquetReader.java

@@ -60,13 +60,15 @@
 
 import static org.apache.hudi.common.util.TypeUtils.unsafeCast;
 import static org.apache.parquet.avro.AvroSchemaConverter.ADD_LIST_ELEMENT_RECORDS;
+import static org.apache.parquet.avro.HoodieAvroParquetSchemaConverter.getAvroSchemaConverter;
 
 public class HoodieSparkParquetReader implements HoodieSparkFileReader {
 
   private final StoragePath path;
   private final HoodieStorage storage;
   private final FileFormatUtils parquetUtils;
   private final List<ClosableIterator> readerIterators = new ArrayList<>();
+  private Option<MessageType> messageTypeOption = Option.empty();
   private Option<StructType> structTypeOption = Option.empty();
   private Option<Schema> schemaOption = Option.empty();
 
@@ -116,19 +118,17 @@ public ClosableIterator<String> getRecordKeyIterator() throws IOException {
   }
 
   public ClosableIterator<UnsafeRow> getUnsafeRowIterator(Schema requestedSchema) throws IOException {
-    return getUnsafeRowIterator(HoodieInternalRowUtils.getCachedSchema(requestedSchema));
-  }
-
-  public ClosableIterator<UnsafeRow> getUnsafeRowIterator(StructType requestedSchema) throws IOException {
-    SparkBasicSchemaEvolution evolution = new SparkBasicSchemaEvolution(getStructSchema(), requestedSchema, SQLConf.get().sessionLocalTimeZone());
+    StructType structSchema = HoodieInternalRowUtils.getCachedSchema(requestedSchema);
+    Option<MessageType> messageSchema = Option.of(getAvroSchemaConverter(storage.getConf().unwrapAs(Configuration.class)).convert(requestedSchema));
+    SparkBasicSchemaEvolution evolution = new SparkBasicSchemaEvolution(getStructSchema(), structSchema, SQLConf.get().sessionLocalTimeZone());
     String readSchemaJson = evolution.getRequestSchema().json();
     storage.getConf().set(ParquetReadSupport.PARQUET_READ_SCHEMA, readSchemaJson);
     storage.getConf().set(ParquetReadSupport.SPARK_ROW_REQUESTED_SCHEMA(), readSchemaJson);
     storage.getConf().set(SQLConf.PARQUET_BINARY_AS_STRING().key(), SQLConf.get().getConf(SQLConf.PARQUET_BINARY_AS_STRING()).toString());
     storage.getConf().set(SQLConf.PARQUET_INT96_AS_TIMESTAMP().key(), SQLConf.get().getConf(SQLConf.PARQUET_INT96_AS_TIMESTAMP()).toString());
     ParquetReader<InternalRow> reader = ParquetReader.builder(new HoodieParquetReadSupport(Option$.MODULE$.empty(), true,
             SparkAdapterSupport$.MODULE$.sparkAdapter().getRebaseSpec("CORRECTED"),
-            SparkAdapterSupport$.MODULE$.sparkAdapter().getRebaseSpec("LEGACY")),
+            SparkAdapterSupport$.MODULE$.sparkAdapter().getRebaseSpec("LEGACY"), messageSchema),
             new Path(path.toUri()))
         .withConf(storage.getConf().unwrapAs(Configuration.class))
         .build();
@@ -139,15 +139,22 @@ public ClosableIterator<UnsafeRow> getUnsafeRowIterator(StructType requestedSche
     return projectedIterator;
   }
 
+  private MessageType getMessageType() {
+    if (messageTypeOption.isEmpty()) {
+      MessageType messageType = ((ParquetUtils) parquetUtils).readSchema(storage, path);
+      messageTypeOption = Option.of(messageType);
+    }
+    return messageTypeOption.get();
+  }
+
   @Override
   public Schema getSchema() {
     if (schemaOption.isEmpty()) {
       // Some types in avro are not compatible with parquet.
       // Avro only supports representing Decimals as fixed byte array
       // and therefore if we convert to Avro directly we'll lose logical type-info.
-      MessageType messageType = ((ParquetUtils) parquetUtils).readSchema(storage, path);
-      StructType structType = new ParquetToSparkSchemaConverter(storage.getConf().unwrapAs(Configuration.class)).convert(messageType);
-      structTypeOption = Option.of(structType);
+      MessageType messageType = getMessageType();
+      StructType structType = getStructSchema();
       schemaOption = Option.of(SparkAdapterSupport$.MODULE$.sparkAdapter()
           .getAvroSchemaConverters()
           .toAvroType(structType, true, messageType.getName(), StringUtils.EMPTY_STRING));
@@ -157,7 +164,9 @@ public Schema getSchema() {
 
   protected StructType getStructSchema() {
     if (structTypeOption.isEmpty()) {
-      getSchema();
+      MessageType messageType = getMessageType();
+      StructType structType = new ParquetToSparkSchemaConverter(storage.getConf().unwrapAs(Configuration.class)).convert(messageType);
+      structTypeOption = Option.of(structType);
     }
     return structTypeOption.get();
   }

hudi-client/hudi-spark-client/src/main/scala/org/apache/hudi/SparkFileFormatInternalRowReaderContext.scala

@@ -33,6 +33,7 @@ import org.apache.hudi.common.util.collection.{CachingIterator, ClosableIterator
 import org.apache.hudi.io.storage.{HoodieSparkFileReaderFactory, HoodieSparkParquetReader}
 import org.apache.hudi.storage.{HoodieStorage, StorageConfiguration, StoragePath}
 import org.apache.hudi.util.CloseableInternalRowIterator
+import org.apache.parquet.avro.AvroSchemaConverter
 import org.apache.spark.sql.HoodieInternalRowUtils
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.catalyst.expressions.JoinedRow
@@ -75,19 +76,28 @@ class SparkFileFormatInternalRowReaderContext(baseFileReader: SparkColumnarFileR
     if (hasRowIndexField) {
       assert(getRecordContext.supportsParquetRowIndex())
     }
-    val structType = HoodieInternalRowUtils.getCachedSchema(requiredSchema)
     if (FSUtils.isLogFile(filePath)) {
       new HoodieSparkFileReaderFactory(storage).newParquetFileReader(filePath)
-        .asInstanceOf[HoodieSparkParquetReader].getUnsafeRowIterator(structType).asInstanceOf[ClosableIterator[InternalRow]]
+        .asInstanceOf[HoodieSparkParquetReader].getUnsafeRowIterator(requiredSchema).asInstanceOf[ClosableIterator[InternalRow]]
     } else {
+      val structType = HoodieInternalRowUtils.getCachedSchema(requiredSchema)
       // partition value is empty because the spark parquet reader will append the partition columns to
       // each row if they are given. That is the only usage of the partition values in the reader.
       val fileInfo = sparkAdapter.getSparkPartitionedFileUtils
         .createPartitionedFile(InternalRow.empty, filePath, start, length)
       val (readSchema, readFilters) = getSchemaAndFiltersForRead(structType, hasRowIndexField)
+
+      // Convert Avro dataSchema to Parquet MessageType for timestamp precision conversion
+      val tableSchemaOpt = if (dataSchema != null) {
+        val hadoopConf = storage.getConf.unwrapAs(classOf[Configuration])
+        val parquetSchema = new AvroSchemaConverter(hadoopConf).convert(dataSchema)
+        org.apache.hudi.common.util.Option.of(parquetSchema)
+      } else {
+        org.apache.hudi.common.util.Option.empty[org.apache.parquet.schema.MessageType]()
+      }
       new CloseableInternalRowIterator(baseFileReader.read(fileInfo,
         readSchema, StructType(Seq.empty), getSchemaHandler.getInternalSchemaOpt,
-        readFilters, storage.getConf.asInstanceOf[StorageConfiguration[Configuration]]))
+        readFilters, storage.getConf.asInstanceOf[StorageConfiguration[Configuration]], tableSchemaOpt))
     }
   }
 

hudi-client/hudi-spark-client/src/main/scala/org/apache/spark/sql/execution/datasources/SparkColumnarFileReader.scala

@@ -23,6 +23,7 @@ import org.apache.hadoop.conf.Configuration
 import org.apache.hudi.common.util
 import org.apache.hudi.internal.schema.InternalSchema
 import org.apache.hudi.storage.StorageConfiguration
+import org.apache.parquet.schema.MessageType
 import org.apache.spark.sql.catalyst.InternalRow
 import org.apache.spark.sql.sources.Filter
 import org.apache.spark.sql.types.StructType
@@ -37,12 +38,14 @@ trait SparkColumnarFileReader extends Serializable {
    * @param internalSchemaOpt  option of internal schema for schema.on.read
    * @param filters            filters for data skipping. Not guaranteed to be used; the spark plan will also apply the filters.
    * @param storageConf        the hadoop conf
+   * @param tableSchemaOpt     option of table schema for timestamp precision conversion
    * @return iterator of rows read from the file output type says [[InternalRow]] but could be [[ColumnarBatch]]
    */
   def read(file: PartitionedFile,
            requiredSchema: StructType,
            partitionSchema: StructType,
            internalSchemaOpt: util.Option[InternalSchema],
            filters: Seq[Filter],
-           storageConf: StorageConfiguration[Configuration]): Iterator[InternalRow]
+           storageConf: StorageConfiguration[Configuration],
+           tableSchemaOpt: util.Option[MessageType] = util.Option.empty()): Iterator[InternalRow]
 }

hudi-client/hudi-spark-client/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/HoodieParquetReadSupport.scala

@@ -24,7 +24,7 @@ import org.apache.hudi.common.util.ValidationUtils
 
 import org.apache.parquet.hadoop.api.InitContext
 import org.apache.parquet.hadoop.api.ReadSupport.ReadContext
-import org.apache.parquet.schema.{GroupType, MessageType, Type, Types}
+import org.apache.parquet.schema.{GroupType, MessageType, SchemaRepair, Type, Types}
 import org.apache.spark.sql.catalyst.util.RebaseDateTime.RebaseSpec
 
 import java.time.ZoneId
@@ -35,16 +35,16 @@ class HoodieParquetReadSupport(
                                 convertTz: Option[ZoneId],
                                 enableVectorizedReader: Boolean,
                                 datetimeRebaseSpec: RebaseSpec,
-                                int96RebaseSpec: RebaseSpec)
+                                int96RebaseSpec: RebaseSpec,
+                                tableSchemaOpt: org.apache.hudi.common.util.Option[org.apache.parquet.schema.MessageType] = org.apache.hudi.common.util.Option.empty())
   extends ParquetReadSupport(convertTz, enableVectorizedReader, datetimeRebaseSpec, int96RebaseSpec) with SparkAdapterSupport {
 
   override def init(context: InitContext): ReadContext = {
     val readContext = super.init(context)
-    val requestedParquetSchema = readContext.getRequestedSchema
+    val requestedParquetSchema = SchemaRepair.repairLogicalTypes(readContext.getRequestedSchema, tableSchemaOpt)
     val trimmedParquetSchema = HoodieParquetReadSupport.trimParquetSchema(requestedParquetSchema, context.getFileSchema)
     new ReadContext(trimmedParquetSchema, readContext.getReadSupportMetadata)
   }
-
 }
 
 object HoodieParquetReadSupport {

hudi-client/hudi-spark-client/src/main/scala/org/apache/spark/sql/hudi/MultipleColumnarFileFormatReader.scala

@@ -43,14 +43,14 @@ class MultipleColumnarFileFormatReader(parquetReader: SparkColumnarFileReader, o
    * @param storageConf       the hadoop conf
    * @return iterator of rows read from the file output type says [[InternalRow]] but could be [[ColumnarBatch]]
    */
-  override def read(file: PartitionedFile, requiredSchema: StructType, partitionSchema: StructType, internalSchemaOpt: util.Option[InternalSchema], filters: Seq[Filter], storageConf: StorageConfiguration[Configuration]): Iterator[InternalRow] = {
+  override def read(file: PartitionedFile, requiredSchema: StructType, partitionSchema: StructType, internalSchemaOpt: util.Option[InternalSchema], filters: Seq[Filter], storageConf: StorageConfiguration[Configuration], tableSchemaOpt: util.Option[org.apache.parquet.schema.MessageType]): Iterator[InternalRow] = {
     val filePath = sparkAdapter.getSparkPartitionedFileUtils.getPathFromPartitionedFile(file)
     val fileFormat = HoodieFileFormat.fromFileExtension(filePath.getFileExtension)
     fileFormat match {
       case HoodieFileFormat.PARQUET =>
-        parquetReader.read(file, requiredSchema, partitionSchema, internalSchemaOpt, filters, storageConf)
+        parquetReader.read(file, requiredSchema, partitionSchema, internalSchemaOpt, filters, storageConf, tableSchemaOpt)
       case HoodieFileFormat.ORC =>
-        orcReader.read(file, requiredSchema, partitionSchema, internalSchemaOpt, filters, storageConf)
+        orcReader.read(file, requiredSchema, partitionSchema, internalSchemaOpt, filters, storageConf, tableSchemaOpt)
       case _ =>
         throw new IllegalArgumentException(s"Unsupported file format for file: $filePath")
     }

hudi-common/src/main/java/org/apache/hudi/avro/HoodieAvroUtils.java

@@ -1208,7 +1208,9 @@ public static Object rewritePrimaryType(Object oldValue, Schema oldSchema, Schem
           return oldValue;
         case LONG:
           if (oldSchema.getLogicalType() != newSchema.getLogicalType()) {
-            if (oldSchema.getLogicalType() instanceof LogicalTypes.TimestampMillis) {
+            if (oldSchema.getLogicalType() == null || newSchema.getLogicalType() == null) {
+              return oldValue;
+            } else if (oldSchema.getLogicalType() instanceof LogicalTypes.TimestampMillis) {
               if (newSchema.getLogicalType() instanceof LogicalTypes.TimestampMicros) {
                 return DateTimeUtils.millisToMicros((Long) oldValue);
               }
@@ -1485,6 +1487,10 @@ public static boolean recordNeedsRewriteForExtendedAvroTypePromotion(Schema writ
       case DOUBLE:
       case FLOAT:
       case LONG:
+        if (readerSchema.getLogicalType() instanceof LogicalTypes.TimestampMillis
+            && writerSchema.getLogicalType() instanceof LogicalTypes.TimestampMicros) {
+          return true;
+        }
         return !(writerSchema.getType().equals(Schema.Type.INT) || writerSchema.getType().equals(Schema.Type.LONG));
       default:
         return !writerSchema.getType().equals(readerSchema.getType());

hudi-common/src/main/java/org/apache/hudi/common/table/log/block/HoodieAvroDataBlock.java

@@ -45,6 +45,7 @@
 import org.apache.avro.io.DecoderFactory;
 import org.apache.avro.io.Encoder;
 import org.apache.avro.io.EncoderFactory;
+import org.apache.parquet.schema.AvroSchemaRepair;
 
 import javax.annotation.Nonnull;
 
@@ -187,11 +188,12 @@ private RecordIterator(Schema readerSchema, Schema writerSchema, byte[] content)
         this.totalRecords = this.dis.readInt();
       }
 
-      if (recordNeedsRewriteForExtendedAvroTypePromotion(writerSchema, readerSchema)) {
-        this.reader = new GenericDatumReader<>(writerSchema, writerSchema);
+      Schema repairedWriterSchema = AvroSchemaRepair.repairLogicalTypes(writerSchema, readerSchema);
+      if (recordNeedsRewriteForExtendedAvroTypePromotion(repairedWriterSchema, readerSchema)) {
+        this.reader = new GenericDatumReader<>(repairedWriterSchema, repairedWriterSchema);
         this.promotedSchema = Option.of(readerSchema);
       } else {
-        this.reader = new GenericDatumReader<>(writerSchema, readerSchema);
+        this.reader = new GenericDatumReader<>(repairedWriterSchema, readerSchema);
       }
     }
 
@@ -272,11 +274,12 @@ private StreamingRecordIterator(Schema readerSchema, Schema writerSchema, Seekab
         this.totalRecords = this.inputStream.readInt();
       }
 
-      if (recordNeedsRewriteForExtendedAvroTypePromotion(writerSchema, readerSchema)) {
-        this.reader = new GenericDatumReader<>(writerSchema, writerSchema);
+      Schema repairedWriterSchema = AvroSchemaRepair.repairLogicalTypes(writerSchema, readerSchema);
+      if (recordNeedsRewriteForExtendedAvroTypePromotion(repairedWriterSchema, readerSchema)) {
+        this.reader = new GenericDatumReader<>(repairedWriterSchema, repairedWriterSchema);
         this.promotedSchema = Option.of(readerSchema);
       } else {
-        this.reader = new GenericDatumReader<>(writerSchema, readerSchema);
+        this.reader = new GenericDatumReader<>(repairedWriterSchema, readerSchema);
       }
 
       this.buffer = ByteBuffer.allocate(Math.min(bufferSize, Math.toIntExact(contentLocation.getBlockSize())));