Github user marmbrus commented on a diff in the pull request:
https://github.com/apache/spark/pull/11709#discussion_r57363361
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/ParquetRelation.scala
---
@@ -269,6 +276,137 @@ private[sql] class DefaultSource extends FileFormat
with DataSourceRegister with
file.getName == ParquetFileWriter.PARQUET_METADATA_FILE
}
+ /**
+ * Returns a function that can be used to read a single file in as an
Iterator of InternalRow.
+ *
+ * @param partitionSchema The schema of the partition column row that
will be present in each
+ * PartitionedFile. These columns should be
prepended to the rows that
+ * are produced by the iterator.
+ * @param dataSchema The schema of the data that should be output for
each row. This may be a
+ * subset of the columns that are present in the file
if column pruning has
+ * occurred.
+ * @param filters A set of filters than can optionally be used to reduce
the number of rows output
+ * @param options A set of string -> string configuration options.
+ * @return
+ */
+ override def buildReader(
+ sqlContext: SQLContext,
+ partitionSchema: StructType,
+ dataSchema: StructType,
+ filters: Seq[Filter],
+ options: Map[String, String]): PartitionedFile =>
Iterator[InternalRow] = {
+ val parquetConf = new
Configuration(sqlContext.sparkContext.hadoopConfiguration)
+ parquetConf.set(ParquetInputFormat.READ_SUPPORT_CLASS,
classOf[CatalystReadSupport].getName)
+ parquetConf.set(
+ CatalystReadSupport.SPARK_ROW_REQUESTED_SCHEMA,
+ CatalystSchemaConverter.checkFieldNames(dataSchema).json)
+ parquetConf.set(
+ CatalystWriteSupport.SPARK_ROW_SCHEMA,
+ CatalystSchemaConverter.checkFieldNames(dataSchema).json)
+
+ // We want to clear this temporary metadata from saving into Parquet
file.
+ // This metadata is only useful for detecting optional columns when
pushdowning filters.
+ val dataSchemaToWrite =
StructType.removeMetadata(StructType.metadataKeyForOptionalField,
+ dataSchema).asInstanceOf[StructType]
+ CatalystWriteSupport.setSchema(dataSchemaToWrite, parquetConf)
+
+ // Sets flags for `CatalystSchemaConverter`
+ parquetConf.setBoolean(
+ SQLConf.PARQUET_BINARY_AS_STRING.key,
+ sqlContext.conf.getConf(SQLConf.PARQUET_BINARY_AS_STRING))
+ parquetConf.setBoolean(
+ SQLConf.PARQUET_INT96_AS_TIMESTAMP.key,
+ sqlContext.conf.getConf(SQLConf.PARQUET_INT96_AS_TIMESTAMP))
+
+ // Try to push down filters when filter push-down is enabled.
+ val pushed = if
(sqlContext.getConf(SQLConf.PARQUET_FILTER_PUSHDOWN_ENABLED.key).toBoolean) {
+ filters
+ // Collects all converted Parquet filter predicates. Notice that
not all predicates can be
+ // converted (`ParquetFilters.createFilter` returns an
`Option`). That's why a `flatMap`
+ // is used here.
+ .flatMap(ParquetFilters.createFilter(dataSchema, _))
+ .reduceOption(FilterApi.and)
+ } else {
+ None
+ }
+
+ val broadcastedConf =
+ sqlContext.sparkContext.broadcast(new
SerializableConfiguration(parquetConf))
+
+ // TODO: if you move this into the closure it reverts to the default
values.
+ // If true, enable using the custom RecordReader for parquet. This
only works for
+ // a subset of the types (no complex types).
+ val enableVectorizedParquetReader: Boolean =
+
sqlContext.getConf(SQLConf.PARQUET_VECTORIZED_READER_ENABLED.key).toBoolean
+ val enableWholestageCodegen: Boolean =
+ sqlContext.getConf(SQLConf.WHOLESTAGE_CODEGEN_ENABLED.key).toBoolean
+
+ (file: PartitionedFile) => {
+ assert(file.partitionValues.numFields == partitionSchema.size)
+
+ val fileSplit =
+ new FileSplit(new Path(new URI(file.filePath)), file.start,
file.length, Array.empty)
+
+ val split =
+ new org.apache.parquet.hadoop.ParquetInputSplit(
+ fileSplit.getPath,
+ fileSplit.getStart,
+ fileSplit.getStart + fileSplit.getLength,
+ fileSplit.getLength,
+ fileSplit.getLocations,
+ null)
+
+ val attemptId = new TaskAttemptID(new TaskID(new JobID(),
TaskType.MAP, 0), 0)
+ val hadoopAttemptContext = new
TaskAttemptContextImpl(broadcastedConf.value.value, attemptId)
+
+ val parquetReader = try {
+ if (!enableVectorizedParquetReader) sys.error("Vectorized reader
turned off.")
+ val vectorizedReader = new VectorizedParquetRecordReader()
+ vectorizedReader.initialize(split, hadoopAttemptContext)
+ logDebug(s"Appending $partitionSchema ${file.partitionValues}")
+ vectorizedReader.initBatch(partitionSchema, file.partitionValues)
--- End diff --
Basically, I'd just take the non-vectorized version below, put it in a
utility function and use it everywhere. If we vectorize all the sources, that
will be the only part we have to remove and then this can be done in
FileScanRDD.
I think that you do not want to do the actually partition appending in the
planner like we were before, because you can't have Spark Partitions (splits)
that read from different partitions very easily. This is what was making the
bucking logic so convoluted in the old code path. This makes bucketing and
collapsing of small files into a single partition much simpler.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org
For additional commands, e-mail: reviews-h...@spark.apache.org
