我有使用小驼峰变量名称约定的案例类。Cassandra 对行名和表名使用lower_underscore约定。有没有办法指示数据帧自动转换名称?
表定义:
CREATE TABLE data_storage.dummy (
some_name text,
some_value text,
PRIMARY KEY (some_name)
);
火花代码:
val ds = List(Dummy("key", 1)).toDS()
ds.write
.format("org.apache.spark.sql.cassandra")
.option("keyspace", "data_storage")
.option("table", "dummy")
.save
我编写了一个将名称从 camel 更改为下划线的函数,反之亦然,但是当某些遗留案例类也使用命名约定的情况下lower_underscore它不起作用。
private def toUnderscore(ds: DataFrame): DataFrame = {
normalizeNames(ds, CaseFormat.LOWER_CAMEL, CaseFormat.LOWER_UNDERSCORE)
}
private def toCamel(ds: DataFrame): DataFrame =
normalizeNames(ds, CaseFormat.LOWER_UNDERSCORE, CaseFormat.LOWER_CAMEL)
private def normalizeNames(df: DataFrame, from: CaseFormat, to: CaseFormat): DataFrame = {
def normalizeName(c: String): String = from.to(to, c.toString)
def flattenSchema(schema: StructType): StructType = {
StructType(schema.fields.map {
case StructField(name, inner: StructType, nullable, metadata) =>
StructField(normalizeName(name), flattenSchema(inner), nullable, metadata)
case StructField(name, array: ArrayType, nullable, metadata) =>
StructField(
normalizeName(name), array.copy(elementType = array.elementType match {
case s: StructType => flattenSchema(s)
case x => x
}), nullable, metadata)
case StructField(name, flat, nullable, metadata) =>
StructField(normalizeName(name), flat, nullable, metadata)
})
}
这种名称转换没有内置方法。RDD 接口有一些围绕其类型转换的其他方法,这些方法执行此类转换,但总的来说,我们没有为数据集/数据帧添加机制。