Writing CSV file using Spark and java - handling empty values and quotes

Question

Initial data is in Dataset<Row> and I am trying to write to pipe delimited file and I want each non empty cell and non null values to be placed in quotes. Empty or null values should not contain quotes

result.coalesce(1).write()
            .option("delimiter", "|")
            .option("header", "true")
            .option("nullValue", "")
            .option("quoteAll", "false")
            .csv(Location);

Expected output:

"London"||"UK"
"Delhi"|"India"
"Moscow"|"Russia"

Current Output:

London||UK
Delhi|India
Moscow|Russia

If I change the "quoteAll" to "true", output I am getting is:

"London"|""|"UK"
"Delhi"|"India"
"Moscow"|"Russia"

Spark version is 2.3 and java version is java 8

Answer 1

Java answer. CSV escape is not just adding " symbols around. You should handle " inside strings. So let's use StringEscapeUtils and define UDF that will call it. Then just apply the UDF to each of the column.

import org.apache.commons.text.StringEscapeUtils;
import org.apache.spark.sql.Column;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import static org.apache.spark.sql.functions.*;
import org.apache.spark.sql.expressions.UserDefinedFunction;
import org.apache.spark.sql.types.DataTypes;

import java.util.Arrays;

public class Test {

    void test(Dataset<Row> result, String Location) {
        // define UDF
        UserDefinedFunction escape = udf(
            (String str) -> str.isEmpty()?"":StringEscapeUtils.escapeCsv(str), DataTypes.StringType
        );
        // call udf for each column
        Column columns[] = Arrays.stream(result.schema().fieldNames())
                .map(f -> escape.apply(col(f)).as(f))
                .toArray(Column[]::new);

         // save the result
        result.select(columns)
                .coalesce(1).write()
                .option("delimiter", "|")
                .option("header", "true")
                .option("nullValue", "")
                .option("quoteAll", "false")
                .csv(Location);
    }
}

Side note: coalesce(1) is a bad call. It collect all data on one executor. You can get executor OOM in production for huge dataset.

Answer 2

EDIT & Warning: Did not see java tag. This is Scala solution that uses foldLeft as a loop to go over all columns. If this is replaced by a Java friendly loop, everything should work as is. I will try and look back at this at the later time.

A programmatic solution could be

val columns = result.columns
val randomColumnName = "RND"

val result2 = columns.foldLeft(result) { (data, column) =>
data
  .withColumnRenamed(column, randomColumnName)
  .withColumn(column,
    when(col(randomColumnName).isNull, "")
      .otherwise(concat(lit("\""), col(randomColumnName), lit("\"")))
  )
  .drop(randomColumnName)
}

This will produce the strings with " around them and write empty strings in nulls. If you need to keep nulls, just keep them.

Then just write it down:

result2.coalesce(1).write()
            .option("delimiter", "|")
            .option("header", "true")
            .option("quoteAll", "false")
            .csv(Location);

Answer 3

This is certainly not a efficient answer and I am modifying this based on one given by Artem Aliev, but thought it would be useful to few people, so posting this answer

import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SparkSession;
import static org.apache.spark.sql.functions.*;<br/>
import org.apache.spark.sql.expressions.UserDefinedFunction;
import org.apache.spark.sql.types.DataTypes;<br/>
public class Quotes {<br/>
    private static final String DELIMITER = "|";
    private static final String Location = "Give location here";

    public static void main(String[] args) {

        SparkSession sparkSession = SparkSession.builder() 
                .master("local") 
                .appName("Spark Session") 
                .enableHiveSupport()
                .getOrCreate();

        Dataset<Row> result = sparkSession.read()
                .option("header", "true")
                .option("delimiter",DELIMITER)
                .csv("Sample file to read"); //Give the details of file to read here

      UserDefinedFunction udfQuotesNonNull = udf(
        (String abc) -> (abc!=null? "\""+abc+"\"":abc),DataTypes.StringType
      );

      result = result.withColumn("ind_val", monotonically_increasing_id()); //inducing a new column to be used for join as there is no identity column in source dataset


      Dataset<Row> dataset1 = result.select((udfQuotesNonNull.apply(col("ind_val").cast("string")).alias("ind_val"))); //Dataset used for storing temporary results
      Dataset<Row> dataset = result.select((udfQuotesNonNull.apply(col("ind_val").cast("string")).alias("ind_val")));  //Dataset used for storing output

      String[] str = result.schema().fieldNames();
      dataset1.show();
      for(int j=0; j<str.length-1;j++)
      {
        dataset1 = result.select((udfQuotesNonNull.apply(col("ind_val").cast("string")).alias("ind_val")),(udfQuotesNonNull.apply(col(str[j]).cast("string")).alias("\""+str[j]+"\""))); 
        dataset=dataset.join(dataset1,"ind_val"); //Joining based on induced column
      }
      result = dataset.drop("ind_val");

      result.coalesce(1).write()
      .option("delimiter", DELIMITER)
      .option("header", "true")
      .option("quoteAll", "false")
      .option("nullValue", null)
      .option("quote", "\u0000") 
      .option("spark.sql.sources.writeJobUUID", false)
      .csv(Location);
    }
}