创建DataFrame，有三种模式，一种是sql()主要是访问Hive表；一种是从RDD生成DataFrame，主要从ExistingRDD开始创建；还有一种是read/format格式，从json/txt/csv等数据源格式创建。
先看看第三种方式的创建流程。
1、read/format
def read: DataFrameReader = new DataFrameReader(self)
SparkSession.read()方法直接创建DataFrameReader，然后再DataFrameReader的load()方法来导入外部数据源。load()方法主要逻辑如下：
def load(paths: String*): DataFrame = {
sparkSession.baseRelationToDataFrame(
DataSource.apply(
sparkSession,
paths = paths,
userSpecifiedSchema = userSpecifiedSchema,
className = source,
options = extraOptions.toMap).resolveRelation())
}创建对应数据源类型的DataSource，DataSource解析成BaseRelation,然后通过SparkSession的baseRelationToDataFrame方法从BaseRelation映射生成DataFrame。从BaseRelation创建LogicalRelation，然后调用Dataset.ofRows方法从LogicalRelation创建DataFrame。DataFrame实际就是Dataset。
type DataFrame = Dataset[Row]
baseRelationToDataFrame的定义：
def baseRelationToDataFrame(baseRelation: BaseRelation): DataFrame = {
Dataset.ofRows(self, LogicalRelation(baseRelation))
}Dataset.ofRows方法主要是将逻辑计划转换成物理计划，然后生成新的Dataset。
2、执行
SparkSession的执行关键是如何从LogicalPlan生成物理计划。我们试试跟踪这部分逻辑。
def count(): Long = withAction("count", groupBy().count().queryExecution) { plan =>
plan.executeCollect().head.getLong(0)
}
Dataset的count()动作触发物理计划的执行，调用物理计划plan的executeCollect方法，该方法实际上会调用doExecute()方法生成Array[InternalRow]格式。executeCollect方法在SparkPlan中定义。
3、HadoopFsRelation
需要跟踪下如何从HadoopFsRelation生成物理计划（也就是SparkPlan）
通过FileSourceStrategy来解析。它在FileSourceScanExec上叠加Filter和Projection等操作，看看FileSourceScanExec的定义：
case class FileSourceScanExec(
@transient relation: HadoopFsRelation,
output: Seq[Attribute],
requiredSchema: StructType,
partitionFilters: Seq[Expression],
dataFilters: Seq[Expression],
override val metastoreTableIdentifier: Option[TableIdentifier])
extends DataSourceScanExec with ColumnarBatchScan {
。。。
}它的主要执行代码doExecute()的功能逻辑如下：
protected override def doExecute(): RDD[InternalRow] = {
if (supportsBatch) {
// in the case of fallback, this batched scan should never fail because of:
// 1) only primitive types are supported
// 2) the number of columns should be smaller than spark.sql.codegen.maxFields
WholeStageCodegenExec(this).execute()
} else {
val unsafeRows = {
val scan = inputRDD
if (needsUnsafeRowConversion) {
scan.mapPartitionsWithIndexInternal { (index, iter) =>
val proj = UnsafeProjection.create(schema)
proj.initialize(index)
iter.map(proj)
}
} else {
scan
}
}
val numOutputRows = longMetric("numOutputRows")
unsafeRows.map { r =>
numOutputRows += 1
r
}
}
}inputRDD有两种方式创建，一是createBucketedReadRDD，二是createNonBucketedReadRDD。两者没有本质的区别，仅仅是文件分区规则的不同。
private lazy val inputRDD: RDD[InternalRow] = {
val readFile: (PartitionedFile) => Iterator[InternalRow] =
relation.fileFormat.buildReaderWithPartitionValues(
sparkSession = relation.sparkSession,
dataSchema = relation.dataSchema,
partitionSchema = relation.partitionSchema,
requiredSchema = requiredSchema,
filters = pushedDownFilters,
options = relation.options,
hadoopConf = relation.sparkSession.sessionState.newHadoopConfWithOptions(relation.options))
relation.bucketSpec match {
case Some(bucketing) if relation.sparkSession.sessionState.conf.bucketingEnabled =>
createBucketedReadRDD(bucketing, readFile, selectedPartitions, relation)
case _ =>
createNonBucketedReadRDD(readFile, selectedPartitions, relation)
}
}
createNonBucketedReadRDD调用FileScanRDD ：
new FileScanRDD(fsRelation.sparkSession, readFile, partitions)