Github user davies commented on a diff in the pull request:
https://github.com/apache/spark/pull/8988#discussion_r41545090
--- Diff:
sql/core/src/main/scala/org/apache/spark/sql/execution/datasources/parquet/CatalystWriteSupport.scala
---
@@ -0,0 +1,432 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one or more
+ * contributor license agreements. See the NOTICE file distributed with
+ * this work for additional information regarding copyright ownership.
+ * The ASF licenses this file to You under the Apache License, Version 2.0
+ * (the "License"); you may not use this file except in compliance with
+ * the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.spark.sql.execution.datasources.parquet
+
+import java.nio.{ByteBuffer, ByteOrder}
+import java.util
+
+import scala.collection.JavaConverters.mapAsJavaMapConverter
+
+import org.apache.hadoop.conf.Configuration
+import org.apache.parquet.column.ParquetProperties
+import org.apache.parquet.hadoop.ParquetOutputFormat
+import org.apache.parquet.hadoop.api.WriteSupport
+import org.apache.parquet.hadoop.api.WriteSupport.WriteContext
+import org.apache.parquet.io.api.{Binary, RecordConsumer}
+
+import org.apache.spark.Logging
+import org.apache.spark.sql.SQLConf
+import org.apache.spark.sql.catalyst.InternalRow
+import org.apache.spark.sql.catalyst.expressions.SpecializedGetters
+import org.apache.spark.sql.catalyst.util.DateTimeUtils
+import
org.apache.spark.sql.execution.datasources.parquet.CatalystSchemaConverter.{MAX_PRECISION_FOR_INT32,
MAX_PRECISION_FOR_INT64, minBytesForPrecision}
+import org.apache.spark.sql.types._
+
+/**
+ * A Parquet [[WriteSupport]] implementation that writes Catalyst
[[InternalRow]]s as Parquet
+ * messages. This class can write Parquet data in two modes:
+ *
+ * - Standard mode: Parquet data are written in standard format defined
in parquet-format spec.
+ * - Legacy mode: Parquet data are written in legacy format compatible
with Spark 1.4 and prior.
+ *
+ * This behavior can be controlled by SQL option
`spark.sql.parquet.writeLegacyFormat`. The value
+ * of this option is propagated to this class by the `init()` method and
its Hadoop configuration
+ * argument.
+ */
+private[parquet] class CatalystWriteSupport extends
WriteSupport[InternalRow] with Logging {
+ // A `ValueWriter` is responsible for writing a field of an
`InternalRow` to the record consumer.
+ // Here we are using `SpecializedGetters` rather than `InternalRow` so
that we can directly access
+ // data in `ArrayData` without the help of `SpecificMutableRow`.
+ private type ValueWriter = (SpecializedGetters, Int) => Unit
+
+ // Schema of the `InternalRow`s to be written
+ private var schema: StructType = _
+
+ // `ValueWriter`s for all fields of the schema
+ private var rootFieldWriters: Seq[ValueWriter] = _
+
+ // The Parquet `RecordConsumer` to which all `InternalRow`s are written
+ private var recordConsumer: RecordConsumer = _
+
+ // Whether to write data in legacy Parquet format compatible with Spark
1.4 and prior versions
+ private var writeLegacyParquetFormat: Boolean = _
+
+ // Reusable byte array used to write timestamps as Parquet INT96 values
+ private val timestampBuffer = new Array[Byte](12)
+
+ // Reusable byte array used to write decimal values
+ private val decimalBuffer = new
Array[Byte](minBytesForPrecision(DecimalType.MAX_PRECISION))
+
+ override def init(configuration: Configuration): WriteContext = {
+ val schemaString =
configuration.get(CatalystWriteSupport.SPARK_ROW_SCHEMA)
+ this.schema = StructType.fromString(schemaString)
+ this.writeLegacyParquetFormat = {
+ // `SQLConf.PARQUET_WRITE_LEGACY_FORMAT` should always be explicitly
set in ParquetRelation
+ assert(configuration.get(SQLConf.PARQUET_WRITE_LEGACY_FORMAT.key) !=
null)
+ configuration.get(SQLConf.PARQUET_WRITE_LEGACY_FORMAT.key).toBoolean
+ }
+ this.rootFieldWriters = schema.map(_.dataType).map(makeWriter)
+
+ val messageType = new
CatalystSchemaConverter(configuration).convert(schema)
+ val metadata = Map(CatalystReadSupport.SPARK_METADATA_KEY ->
schemaString).asJava
+
+ logInfo(
+ s"""Initialized Parquet WriteSupport with Catalyst schema:
+ |${schema.prettyJson}
+ |and corresponding Parquet message type:
+ |$messageType
+ """.stripMargin)
+
+ new WriteContext(messageType, metadata)
+ }
+
+ override def prepareForWrite(recordConsumer: RecordConsumer): Unit = {
+ this.recordConsumer = recordConsumer
+ }
+
+ override def write(row: InternalRow): Unit = {
+ consumeMessage(writeFields(row, schema, rootFieldWriters))
+ }
+
+ private def writeFields(
+ row: InternalRow, schema: StructType, fieldWriters:
Seq[ValueWriter]): Unit = {
+ var i = 0
+ while (i < row.numFields) {
+ if (!row.isNullAt(i)) {
+ consumeField(schema(i).name, i) {
+ fieldWriters(i).apply(row, i)
+ }
+ }
+ i += 1
+ }
+ }
+
+ private def makeWriter(dataType: DataType): ValueWriter = {
+ dataType match {
+ case BooleanType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addBoolean(row.getBoolean(ordinal))
+
+ case ByteType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addInteger(row.getByte(ordinal))
+
+ case ShortType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addInteger(row.getShort(ordinal))
+
+ case IntegerType | DateType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addInteger(row.getInt(ordinal))
+
+ case LongType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addLong(row.getLong(ordinal))
+
+ case FloatType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addFloat(row.getFloat(ordinal))
+
+ case DoubleType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+ recordConsumer.addDouble(row.getDouble(ordinal))
+
+ case StringType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+
recordConsumer.addBinary(Binary.fromByteArray(row.getUTF8String(ordinal).getBytes))
+
+ case TimestampType =>
+ (row: SpecializedGetters, ordinal: Int) => {
+ // TODO Writes `TimestampType` values as `TIMESTAMP_MICROS` once
parquet-mr implements it
+ // Currently we only support timestamps stored as INT96, which
is compatible with Hive
+ // and Impala. However, INT96 is to be deprecated. We plan to
support `TIMESTAMP_MICROS`
+ // defined in the parquet-format spec. But up until writing,
the most recent parquet-mr
+ // version (1.8.1) hasn't implemented it yet.
+
+ // NOTE: Starting from Spark 1.5, Spark SQL `TimestampType` only
has microsecond
+ // precision. Nanosecond parts of timestamp values read from
INT96 are simply stripped.
+ val (julianDay, timeOfDayNanos) =
DateTimeUtils.toJulianDay(row.getLong(ordinal))
+ val buf = ByteBuffer.wrap(timestampBuffer)
+
buf.order(ByteOrder.LITTLE_ENDIAN).putLong(timeOfDayNanos).putInt(julianDay)
+ recordConsumer.addBinary(Binary.fromByteArray(timestampBuffer))
+ }
+
+ case BinaryType =>
+ (row: SpecializedGetters, ordinal: Int) =>
+
recordConsumer.addBinary(Binary.fromByteArray(row.getBinary(ordinal)))
+
+ case DecimalType.Fixed(precision, scale) =>
+ makeDecimalWriter(precision, scale)
+
+ case t: StructType =>
+ val fieldWriters = t.map(_.dataType).map(makeWriter)
+ (row: SpecializedGetters, ordinal: Int) =>
+ consumeGroup(writeFields(row.getStruct(ordinal, t.length), t,
fieldWriters))
--- End diff --
```
consumeGroup {
}
```
---
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
