Spark UDF类型不匹配错误

Question

I'm trying to write a UDF to convert a timestamp into an integer representing the hour of the week. I'm easily able to accomplish this with SparkSql like this.

I have many UDFs in our code with this exact syntax but this one is trying a type mismatch error. I also tried invoking my UDF with col("session_ts_start") but that also failed to work.

import spark.implicits._
import java.sql.Timestamp
import org.apache.spark.sql.functions._

def getHourOfWeek() = udf(
    (ts: Timestamp) => unix_timestamp(ts)
)

val dDF = df.withColumn("hour", getHourOfWeek()(df("session_ts_start")))
dDF.show()

<console>:154: error: type mismatch;
 found   : java.sql.Timestamp
 required: org.apache.spark.sql.Column
           (ts: Timestamp) => unix_timestamp(ts)

Answer 1

unix_timestamp is a SQL function. It operates on Columns not external values:

def unix_timestamp(s: Column): Column 

and it cannot be used in UDF.

I'm trying (...) to convert a timestamp into an integer representing the hour of the week

import org.apache.spark.sql.Column
import org.apache.spark.sql.functions.{date_format, hour}

def getHourOfWeek(c: Column) =
  // https://docs.oracle.com/javase/8/docs/api/java/text/SimpleDateFormat.html
  (date_format(c, "u").cast("integer") - 1) * 24 + hour(c)

val df = Seq("2017-03-07 01:00:00").toDF("ts").select($"ts".cast("timestamp"))

df.select(getHourOfWeek($"ts").alias("hour")).show

+----+
|hour|
+----+
|  25|
+----+

Another possible solution:

import org.apache.spark.sql.functions.{next_day, date_sub}

def getHourOfWeek2(c: Column) = ((
  c.cast("bigint") - 
  date_sub(next_day(c, "Mon"), 7).cast("timestamp").cast("bigint")
) / 3600).cast("int")

df.select(getHourOfWeek2($"ts").alias("hour"))

+----+
|hour|
+----+
|  25|
+----+

Note : Neither solution handles daylight saving time or other date / time subtleties.