我不确定这是一个错误（或者只是语法不正确）。我四处搜索，没有在其他地方看到这一点，所以我在提交错误报告之前在这里询问。
我正在尝试使用在嵌套列上分区的 Window 函数。我在下面创建了一个小示例来演示这个问题。
import sqlContext.implicits._
import org.apache.spark.sql.functions._
import org.apache.spark.sql.expressions.Window
val data = Seq(("a", "b", "c", 3), ("c", "b", "a", 3)).toDF("A", "B", "C", "num")
.withColumn("Data", struct("A", "B", "C")).drop("A").drop("B").drop("C")
val winSpec = Window.partitionBy("Data.A", "Data.B").orderBy($"num".desc)
data.select($"*", max("num").over(winSpec) as "max").where("num = max").drop("max").show
以上导致错误 org.apache.spark.sql.AnalysisException: resolved attribute(s) A#39,B#40 missing from num#33,Data#37 in operator !Project [num#33,Data#37,A#39,B#40];
at org.apache.spark.sql.catalyst.analysis.CheckAnalysis$class.failAnalysis(CheckAnalysis.scala:38)
at org.apache.spark.sql.catalyst.analysis.Analyzer.failAnalysis(Analyzer.scala:44)
...
相反，如果这些列不是嵌套的，则工作正常。我是否缺少语法的某些内容，或者这是一个错误？
在我看来，当
分析器尝试扩展*时，您遇到了一个错误
import sqlContext.implicits._
import org.apache.spark.sql.functions._
import org.apache.spark.sql.expressions.Window
sql("SET spark.sql.eagerAnalysis=false") // Let us see the error even though we are constructing an invalid tree
val data = Seq(("a", "b", "c", 3), ("c", "b", "a", 3)).toDF("A", "B", "C", "num")
.withColumn("Data", struct("A", "B", "C"))
.drop("A")
.drop("B")
.drop("C")
val winSpec = Window.partitionBy("Data.A", "Data.B").orderBy($"num".desc)
data.select($"*", max("num").over(winSpec) as "max").explain(true)
通过关闭预先分析（以便我们可以调用 explain 而不会引发错误），您可以看到"*"正在扩展以包含实际上不可用的列：
== Parsed Logical Plan ==
'Project [*,'max('num) windowspecdefinition('Data.A,'Data.B,'num DESC,UnspecifiedFrame) AS max#64928]
+- Project [num#64926,Data#64927]
+- Project [C#64925,num#64926,Data#64927]
+- Project [B#64924,C#64925,num#64926,Data#64927]
+- Project [A#64923,B#64924,C#64925,num#64926,struct(A#64923,B#64924,C#64925) AS Data#64927]
+- Project [_1#64919 AS A#64923,_2#64920 AS B#64924,_3#64921 AS C#64925,_4#64922 AS num#64926]
+- LocalRelation [_1#64919,_2#64920,_3#64921,_4#64922], [[a,b,c,3],[c,b,a,3]]
== Analyzed Logical Plan ==
num: int, Data: struct<A:string,B:string,C:string>, max: int
Project [num#64926,Data#64927,max#64928]
+- Project [num#64926,Data#64927,A#64932,B#64933,max#64928,max#64928]
+- Window [num#64926,Data#64927,A#64932,B#64933], [HiveWindowFunction#org.apache.hadoop.hive.ql.udf.generic.GenericUDAFMax(num#64926) windowspecdefinition(A#64932,B#64933,num#64926 DESC,RANGE BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS max#64928], [A#64932,B#64933], [num#64926 DESC]
+- !Project [num#64926,Data#64927,A#64932,B#64933]
+- Project [num#64926,Data#64927]
+- Project [C#64925,num#64926,Data#64927]
+- Project [B#64924,C#64925,num#64926,Data#64927]
+- Project [A#64923,B#64924,C#64925,num#64926,struct(A#64923,B#64924,C#64925) AS Data#64927]
+- Project [_1#64919 AS A#64923,_2#64920 AS B#64924,_3#64921 AS C#64925,_4#64922 AS num#64926]
+- LocalRelation [_1#64919,_2#64920,_3#64921,_4#64922], [[a,b,c,3],[c,b,a,3]]
我在这里提交了这个：https://issues.apache.org/jira/browse/SPARK-12989。如果您手动列出列而不是使用应充当解决方法的*。