spark全局计数器每次执行结果不一样

最近闲着无聊从网上下载了一个去年叫流行的2000W开房数据，数据格式如下：

基本数据格式

对其中一个文件进行解析，解析的时候发现gender那一栏有时候没有数据，然后就想着如果gender一栏数据不符合要求就从身份证号对性别进行解析，我想统计符合要求的记录和不符合要求的记录，用了两个计数器进行计算，代码如下：

import java.sql.Timestamp
import java.text.SimpleDateFormat

import org.apache.spark.sql.SaveMode
import org.apache.spark.{Accumulator, SparkConf, SparkContext}
case class KFCustomer2(name: String, gender: String, ctfId: String, birthday: String,
                  address: String, constellation: String)

object AnalysisKF2 {

  val filePath = "/home/panda/2000W/last5000.csv"
  val conf = new SparkConf().setAppName("AnalysisKF").setMaster("local[3]")
  val sc = new SparkContext(conf)
  val useRec = sc.accumulator(0, "useful Records")
  val uselessRec = sc.accumulator(0, "useless Records")

  def main(args: Array[String]) {

sc.setLogLevel("ERROR")
val sqlContext = new org.apache.spark.sql.SQLContext(sc)
import sqlContext.implicits._

val customer = sc.textFile(filePath).map(_.split(",")).filter(line => line.length > 7).filter(!_.contains("Name")).map(
   p => KFCustomer2(p(0), formatGender(p(4), p(5)), p(4), p(6), p(7), p(6))).toDF()
customer.registerTempTable("customer")
var result = sqlContext.sql("SELECT gender, count(gender) FROM customer where gender = 'F' or gender = 'M' group by gender")
result.collect().foreach(println)

println("************************************************")
println(useRec.name.get + " " + useRec.value)

println(uselessRec.name.get + " " + uselessRec.value)
println("************************************************")

  }

  def formatGender(ctfId: String, gender: String): String = {
var rt = "未知"
if (gender.length != 1) {
   if (ctfId.length == 18) {
      try {
                           if ((ctfId.substring(16, 17).toInt) % 2 == 0)                            {
                                 rt = "M"
                           }
                           else {
                                 rt = "F"
                           }
                        useRec.add(1)
         }
         catch

         {
               case _=> println("error ctfid is : " + ctfId)
               uselessRec.add(1)
         }

   }
else
   {
      uselessRec.add(1)
   }
}
else
{
   useRec.add(1)
   rt = gender
}
rt
  }

}

运行两次Accumulator计算结果都不一样：
[F,17091]
[M,32239]
************************************************
useful Records 49777
useless Records 223
************************************************

[F,17091]
[M,32239]
************************************************
useful Records 49454
useless Records 223
************************************************

不知道什么原因，求高人指点