Field data validation using spark dataframe

I have not really worked with Spark 2.0, so I'll try answering your question with a solution in Spark 1.6.

 // Load you base data
val input  = <<you input dataframe>>

//Extract the schema of your base data
val originalSchema = input.schema

// Modify you existing schema with you additional metadata fields
val modifiedSchema= originalSchema.add("ERROR_COLUMN", StringType, true)
                                  .add("ERROR_VALUE", StringType, true)
                                  .add("ERROR_DESCRIPTION", StringType, true)

// write a custom validation function                                 
def validateColumns(row: Row): Row = {

var err_col: String = null
var err_val: String = null
var err_desc: String = null
val empId = row.getAs[String]("EMPID")
val ename = row.getAs[String]("ENAME")
val gender = row.getAs[String]("GENDER")

// do checking here and populate (err_col,err_val,err_desc) with values if applicable

Row.merge(row, Row(err_col),Row(err_val),Row(err_desc))
}

// Call you custom validation function
val validateDF = input.map { row => validateColumns(row) }  

// Reconstruct the DataFrame with additional columns                      
val checkedDf = sqlContext.createDataFrame(validateDF, newSchema)

// Filter out row having errors
val errorDf = checkedDf.filter($"ERROR_COLUMN".isNotNull && $"ERROR_VALUE".isNotNull && $"ERROR_DESCRIPTION".isNotNull)

// Filter our row having no errors
val errorFreeDf = checkedDf.filter($"ERROR_COLUMN".isNull && !$"ERROR_VALUE".isNull && !$"ERROR_DESCRIPTION".isNull)

I have used this approach personally and it works for me. I hope it points you in the right direction.