问题描述：

spark程序数据量不大，可是计算很多，计算完的新字段都join到原来的dataset中，又反复依靠新的结果计算，反复join到原dataset,这就导致了又大量的join，rdd依赖变得异常复杂而且依赖也很长

spark程序计算代码如下

df=df.withColumn("IS_PASS_SUBJECT",when($"SUBJECT_SCORE"/$"FULL_SCORE" >=0.6,1).otherwise(0))
....大量的添加新列（上百个）

df=df.join(
  df.where("SCHOOL_RANK <= 10").groupBy("SCHOOL_ID","SUBJECT_ID").agg(round(avg("SUBJECT_SCORE"),2).as("SCHOOL_TOP10_SUBJECT_SCORE_AVG")),
  Seq("SCHOOL_ID","SUBJECT_ID"),"left"
)
....大量的计算新字段join到原来df

df=df.withColumn("SINGLE_SCORE_PROJECT_RANK", rank().over(Window.partitionBy($"PROJECT_ID",$"SUBJECT_ID").orderBy($"SUBJECT_SCORE".desc)))
...大量的根据前面计算的新列来计算新列

df=df.join(
  df.where("CLASS_RANK <= 10").groupBy("CLASS_ID","SUBJECT_ID").agg(round(avg("SUBJECT_SCORE"),2).as("CLASS_TOP10_SUBJECT_SCORE_AVG")),
  Seq("CLASS_ID","SUBJECT_ID"),"left"
)
....大量的根据前面计算的结果来计算新的结果然后join到原df

这样的话，spark程序就会在中间卡住，什么都不运行，也不报任何错误,也不重新划分job和stage,什么都卡住不动

我也知道如果没有action算子，spark会根据计算链条中的算子来划分stage，我的这个计算链条太长，我也不知道如何来解决卡死的问题？