select / drop does not really drop the column? - scala

Question

I think I don't undestand how select or drop are working.
I am exploding a dataset and I don't want some of the columns to be copied to the newly generated entries.
val ds = spark.sparkContext.parallelize(Seq(
("2017-01-01 06:15:00", "ASC_a", "1"),
("2017-01-01 06:19:00", "start", "2"),
("2017-01-01 06:22:00", "ASC_b", "2"),
("2017-01-01 06:30:00", "end", "2"),
("2017-01-01 10:45:00", "ASC_a", "3"),
("2017-01-01 10:50:00", "start", "3"),
("2017-01-01 11:22:00", "ASC_c", "4"),
("2017-01-01 11:31:00", "end", "5" )
)).toDF("timestamp", "status", "msg")
ds.show()
val foo = ds.select($"timestamp", $"msg")
val bar = ds.drop($"status")
foo.printSchema()
bar.printSchema()
println("foo " + foo.where($"status" === "end").count)
println("bar" + bar.where($"status" === "end").count)
Output:
root
|-- timestamp: string (nullable = true)
|-- msg: string (nullable = true)
root
|-- timestamp: string (nullable = true)
|-- msg: string (nullable = true)
foo 2
bar 2
Why do I still get an output of 2 for both though I
a) did not select status
b) dropped status
EDIT:
println("foo " + foo.where(foo.col("status") === "end").count) says that there is no column status. Should this not be the same as println("foo " + foo.where($"status" === "end").count)?

Answer 1

Why do I still get an output of 2 for both
Because optimizer is free to reorganize the execution plan. In fact if you check it:
== Physical Plan ==
*Project [_1#4 AS timestamp#8, _3#6 AS msg#10]
+- *Filter (isnotnull(_2#5) && (_2#5 = end))
+- *SerializeFromObject [staticinvoke(class org.apache.spark.unsafe.types.UTF8String, StringType, fromString, assertnotnull(input[0, scala.Tuple3, true])._1, true) AS _1#4, staticinvoke(class org.apache.spark.unsafe.types.UTF8String, StringType, fromString, assertnotnull(input[0, scala.Tuple3, true])._2, true) AS _2#5, staticinvoke(class org.apache.spark.unsafe.types.UTF8String, StringType, fromString, assertnotnull(input[0, scala.Tuple3, true])._3, true) AS _3#6]
+- Scan ExternalRDDScan[obj#3]
you'll see that filter is pushed down as early as possible and executed before project. So it is equivalent to:
SELECT _1 AS timetsatmp, _2 AS msg
FROM ds WHERE _2 IS NOT NULL AND _2 = 'end'
Arguably it is a minor bug, and code should be translated as
SELECT * FROM (
SELECT _1 AS timetsatmp, _2 AS msg FROM ds
) WHERE _2 IS NOT NULL AND _2 = 'end'
and throw an exception.