Question

当我尝试在Spark 1.4.1上执行以下操作时，我得到org.apache.spark.SparkException: Task not serializable：

import java.sql.{Date, Timestamp}
import java.text.SimpleDateFormat

object ConversionUtils {
  val iso8601 = new SimpleDateFormat("yyyy-MM-dd'T'HH:mm:ss.SSSX")

  def tsUTC(s: String): Timestamp = new Timestamp(iso8601.parse(s).getTime)

  val castTS = udf[Timestamp, String](tsUTC _)
}

val df = frame.withColumn("ts", ConversionUtils.castTS(frame("ts_str")))
df.first

此处，frame是DataFrame，位于HiveContext内。该数据框没有任何问题。

我有类似的整数UDF，它们没有任何问题。但是，带有时间戳的那个似乎会引起问题。根据{{3}}，java.sql.TimeStamp实现Serializable，所以这不是问题所在。 SimpleDateFormat也是如此documentation。

这让我相信这是导致问题的UDF。但是，我不确定是什么以及如何解决它。

追踪的相关部分：

Caused by: java.io.NotSerializableException: ...
Serialization stack:
        - object not serializable (class: ..., value: ...$ConversionUtils$@63ed11dd)
        - field (class: ...$ConversionUtils$$anonfun$3, name: $outer, type: class ...$ConversionUtils$)
        - object (class ...$ConversionUtils$$anonfun$3, <function1>)
        - field (class: org.apache.spark.sql.catalyst.expressions.ScalaUdf$$anonfun$2, name: func$2, type: interface scala.Function1)
        - object (class org.apache.spark.sql.catalyst.expressions.ScalaUdf$$anonfun$2, <function1>)
        - field (class: org.apache.spark.sql.catalyst.expressions.ScalaUdf, name: f, type: interface scala.Function1)
        - object (class org.apache.spark.sql.catalyst.expressions.ScalaUdf, scalaUDF(ts_str#2683))
        - field (class: org.apache.spark.sql.catalyst.expressions.Alias, name: child, type: class org.apache.spark.sql.catalyst.expressions.Expression)
        - object (class org.apache.spark.sql.catalyst.expressions.Alias, scalaUDF(ts_str#2683) AS ts#7146)
        - element of array (index: 35)
        - array (class [Ljava.lang.Object;, size 36)
        - field (class: scala.collection.mutable.ArrayBuffer, name: array, type: class [Ljava.lang.Object;)
        - object (class scala.collection.mutable.ArrayBuffer,

Answer 1

尝试：

object ConversionUtils extends Serializable {
  ...
}

Spark：DataFrame上的UDF任务不可序列化

1 个答案: