I am writing a program that reads from DB and outputs to a CSV file. Besides the regular columnar data there are 2 JSON fields data as well. The table layout looks like this (other fields removed for brevity):

| ID | Product_Type | Json_Data   | Demographic_Questions |
| 1  | DPI          | {some_JSON} | {another_JSON}        |
| 2  | Travel       | {some_JSON} | {another_JSON}        |

Program logic

  1. Read data from DB
  2. Store columnar data into a map
  3. Convert JSON data into CSV format and store into the map
  4. Write map into CSV file

The main program

public static void extractData(String lastRunDateTime, String extractionType) throws Exception
    List<LinkedHashMap<String, String>> flatJson = new ArrayList<LinkedHashMap<String, String>>();
    String result = "";
    ResultSet rsData = null;

    List<String> productType = new ArrayList<>(); // Store Product Type name for SQL & CSV creation

    try {
        conn = dbUtil.dbConnect();

        String sqlQuery = "SELECT DISTINCT Product_Type FROM Mapping WHERE Extraction_Type = '"+ extractionType +"'";
        st = conn.prepareStatement(sqlQuery);
        rsData = st.executeQuery();
        while(rsData.next()) {

        // Currently there are 4 product types in DB
        for(int i = 0; i < productType.size(); i++)
            isEmptyRS = true;

            String sProductType = productType.get(i);

            LOG.debug("Extraction Started for WebDB data " + sProductType + " V" + extractionType);

            sqlQuery =
                "SELECT b.Rate, b.Comments, a.* " +
                "FROM users_data a LEFT OUTER JOIN users_ratings b ON a.sys_policy_no = b.sys_policy_no " +
                "WHERE a.Product_Type = '" + sProductType + "' " + 
                "AND CAST(a.Submitted_Date as date) BETWEEN '2016-08-01 00:00:00' AND '" + currentDateTime + "' " +
                "ORDER BY a.Submitted_Date DESC";

            st = conn.prepareStatement(sqlQuery);
            rsData = st.executeQuery();

                LinkedHashMap<String, String> map = new LinkedHashMap<String, String>();

                //LOG.debug("Sys_Policy_No = " + rsData.getString("Sys_Policy_No"));

                map.put("ID", rsData.getString("ID"));
                map.put("Product_type", rsData.getString("Product_type"));

                // Read JSON data and convert to columns
                result = rsData.getString("Json_Data");
                if(result != null && result.length() != 0) {
                    addKeys("", new ObjectMapper().readTree(result), map);

                // Read Demographic data and convert to columns
                result = rsData.getString("Demographic_Questions");
                if(result != null && result.length() != 0) {
                    addKeys("", new ObjectMapper().readTree(result), map);


            String filepath = config.getPropValue("GetAllJsonDataFilePath");
            String filename = filepath + sProductType + "_csv_json_all_V" + extractionType + ".csv";
            LOG.debug("Writing " + filename + "...");

            CSVWriter writer = new CSVWriter();
            writer.writeAsCSV(flatJson , filename);

            LOG.debug("Extraction Completed for WebDB data " + sProductType + " V" + extractionType + "\n");
    } catch (Exception e) {
        LOG.error(e.getMessage(), e);
    } finally {
        if(conn!=null) conn.close();

JSON conversion program

private static void addKeys(String currentPath, JsonNode jsonNode, Map<String, String> map) {
        if (jsonNode.isObject()) {
            ObjectNode objectNode = (ObjectNode) jsonNode;
            Iterator<Map.Entry<String, JsonNode>> iter = objectNode.fields();
            String pathPrefix = currentPath.isEmpty() ? "" : currentPath + ".";

            while (iter.hasNext()) {
                Map.Entry<String, JsonNode> entry = iter.next();
                addKeys(pathPrefix + entry.getKey(), entry.getValue(), map);
        } else if (jsonNode.isArray()) {
            ArrayNode arrayNode = (ArrayNode) jsonNode;
            for (int i = 0; i < arrayNode.size(); i++) {
                addKeys(currentPath + "_" + i, arrayNode.get(i), map);
        } else if (jsonNode.isValueNode()) {
            ValueNode valueNode = (ValueNode) jsonNode;
            String value = valueNode.asText().replace("\n", ". ").replace("\r", "");
            map.put(currentPath, value);

CSV Writer program

public void writeAsCSV(List<LinkedHashMap<String, String>> flatJson, String fileName) throws IOException {
        LinkedHashSet<String> headers = collectHeadersOrdered(flatJson);
        String output = StringUtils.join(headers.toArray(), ",") + "\n";
        for (LinkedHashMap<String, String> linkedMap : flatJson) {
            output = output + getCommaSeparatedRow(headers, linkedMap) + "\n";
        writeToFile(output, fileName);

private LinkedHashSet<String> collectHeadersOrdered(List<LinkedHashMap<String, String>> flatJson) {
        LinkedHashSet<String> headers = new LinkedHashSet<String>();
        for (LinkedHashMap<String, String> linkedMap : flatJson) {
        return headers;

private String getCommaSeparatedRow(Set<String> headers, Map<String, String> map) {
        List<String> items = new ArrayList<String>();
        for (String header : headers) {
            String value = map.get(header) == null ? "" : map.get(header).replace(",", "");
        return StringUtils.join(items.toArray(), ",");

private void writeToFile(String output, String fileName) throws IOException {
    try (BufferedWriter bw =
            new BufferedWriter(new FileWriter(fileName))) {
        LOG.debug("Generating " + fileName + " ...");
    } catch (IOException e) {
        LOG.error(e.getMessage(), e);

While the program runs fine without error it is taking way too much time to execute, roughly 3-4 hours. Currently, the biggest CSV filesize is at 40 MB (around 200k rows, 1300 columns). The 2 JSON fields are subjected to very frequent change and I've seen it growing by 30 data elements every other months.

What can I do to increase the performance?


1 Answer 1


JSON library

Assuming ObjectMapper is from the Jackson library, I think you should be able to create only one instance of it as it's safe to do so. Pro-tip: on that link, the developer of Jackson also suggests using ObjectReader/ObjectWriter if you are using Jackson 2.x.


Since Java 7, you can use try-with-resources to safely and efficiently manage I/O resources, such as your JDBC-related resources. More specifically, you can take a look at this WebLogic blog article to better understand how you can use it for the Connection, Statement and ResultSet objects together.

Variable scope

Your flatJson List is declared quite early on, necessitating you to keep clear()-ing it for each iteration. You can instead considering creating a new List each time.

Modeling JSON as a domain object (...?)

This is just a thought, how about modeling the JSON as a domain object, so that you do less of addKeys() yourself, and perhaps just need a nice toCsvMap() implementation on the domain object to get the Map output you need?

Of course, this very much depends on what you mean 'growing by 30 data elements every other months' as... are these elements just part of an array that your JSON library can easily output as a List? Or do you really mean the JSON payload mutates in different ways even between rows, such that there's no one coherent structure to map it as an object?

SQL Server 2016?

If you are using SQL Server 2016, it looks like you may also rely on it to convert your JSON data to rows and columns... again, per disclaimer above, this depends on how its structure changes over time.

Optimizing bottlenecks

Last but not least, have you already tried profiling - regardless of using precise instrumentation frameworks, or just informally with a stopwatch - your application from the time it queries the database to the time the CSV file is generated? Can the database query be further optimized? Is there some inherent network latency somewhere that is making the code appear to work slower? Is writer.writeAsCSV(flatJson , filename) reasonably efficient? See below.

Writing output

Instead of doing a sub-optimal String concatenation using + in each iteration, consider using the newer Files.write(Path, Iterable, CharSet, OpenOption) method. You just need to map each LinkedHashMap element of your List to a String, and the method will iterate through them for you. In addition, it uses the OS-specific line separator, which may be more preferable depending on your use case.

Since you have 1300 columns, performing a Map.get(Object) twice to retrieve the value for each column, per row, is not going to be the fastest way to do so. Just hold on to that thought for a moment...

Think of the children consumers!

Actually, why is there this requirement to write such a 'sparse' CSV file, where there is never a complete row, and instead you are going to have blocks of values, and then blocks of emptiness depending on the product?

I suppose the output will resemble something the following, if we can sort the rows by product type and there are no other overlapping columns other than the ID and product type:


Will it not be better off to create one CSV file per product type, so that the consumers of these data can fully process the product-type-specific file they require, instead of having to cherry-pick columns from a 40 MB file, which will likely be slower as well?

Writing output (cont'd)

Resuming from the earlier section, Java 8 has a Map.getOrDefault(Object, V) method that simplifies your approach of calling Map.get(Object) twice:

// String value = map.get(header) == null ? "" : map.get(header).replace(",", "");
String value = map.getOrDefault(header, "").replace(",", "");

The conversion of a List<LinkedHashMap<String, String> to a List<String> is relatively straightforward when you think of the approach as such:

  1. Create a map of the total columns you have, with elements mapping to themselves, and treat this as the zeroth row, i.e. a single-element List<Map<String, String>>.

  2. Create a stream out of the zeroth row and your actual payload (\$1...n\$ rows), so that you can apply the common step of mapping each column header against all the \$n + 1\$ Maps and concatenating them as a String.

Putting it altogether:

private static List<String> flattenAll(List<LinkedHashMap<String, String>> input) {
    Set<String> columns = input.stream()
                                .flatMap(v -> v.keySet().stream())
    Map<String, String> header = columns.stream()
                                        .collect(Collectors.toMap(k -> k, v -> v));
    return Stream.concat(Stream.of(header), input.stream())
                .map(m -> columns.stream()
                                .map(k -> m.getOrDefault(k, "").replace(",", ""))

How we get our columns is similar to your original approach but done with a stream-based approach, i.e. to flatMap() each Map.keySet() into a Stream<String>, before collect()-ing them into a LinkedHashSet.

  • \$\begingroup\$ I am not able to create the JSON as a domain object because the table is used by different product type (also means different software vendors). Due to this, the JSON is vastly different. For example with DPI product, it is saved as Customer_ID but in Travel product it is saved as PolicyholderID. Also some products do not make use of Demographic_Questions. I'll try the other suggestions and update the outcome. \$\endgroup\$
    – Maruli
    Dec 7, 2016 at 23:59
  • 1
    \$\begingroup\$ I've implemented the first three suggestions and noticed slight improvement which makes me think the bottleneck might be in the CSV writer part. I have included snapshot of the CSV writer in my post. \$\endgroup\$
    – Maruli
    Dec 8, 2016 at 9:56
  • \$\begingroup\$ @Maruli updated my answer. \$\endgroup\$
    – h.j.k.
    Dec 8, 2016 at 17:10
  • \$\begingroup\$ You hit the bullseye! Currently the program produces CSV for each product type. This makes more sense for user who is analysing the data. The 40 MB file I mentioned earlier is regarding to 1 particular product type (i.e. Travel) only. The other product types are smaller in size, ranging from 1 MB to 20 MB. \$\endgroup\$
    – Maruli
    Dec 9, 2016 at 2:06
  • 1
    \$\begingroup\$ Thank you so much for your guidance. I just want to let you know it is taking less than 10 minutes to execute now, as opposed to 4 hours previously!!! \$\endgroup\$
    – Maruli
    Dec 15, 2016 at 9:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.