Question

Apologies for the long code post but am wondering if someone can help with a multithreading question (I am quite new to multi-threading). I am trying to design a facade class to a RESTFUL web services API that can be shared with multiple threads. I am using HttpURLConnection to do the connection and Google GSON to convert to and from JSON data.

The below class is what I have so far. In this example it has one public method to make an API call (authenticateCustomer()) and the private methods are used to facilitate the API call (i.e to build the POST data string, make a POST request etc).

I make one instance of this class and share it with 1000 threads. The threads call the authenticateCustomer() method. Most of the threads work but there is some threads that get a null pointer exception which is because I haven’t implemented any synchronization. If I make the authenticateCustomer() method ‘synchronized’ it works. The problem is this results in poor concurrency (say, for example, the POST request suddenly takes a long time to complete, this will then hold up all the other threads).

Now to my question. Is the below class not stateless and therefore thread-safe? The very few fields that are in the class are declared final and assigned in the constructor. All of the methods use local variables. The Gson object is stateless (according to their web site) and created as a local variable in the API method anyway.

public final class QuizSyncAPIFacade 
{
    // API Connection Details
private final String m_apiDomain;
private final String m_apiContentType;
private final int m_bufferSize;

// Constructors
public QuizSyncAPIFacade()
{
    m_apiDomain      = "http://*****************************";
    m_apiContentType = ".json";
    m_bufferSize = 8192; // 8k
}

private String readInputStream(InputStream stream) throws IOException
{
        // Create a buffer for the input stream
    byte[] buffer = new byte[m_bufferSize];

    int readCount;

    StringBuilder builder = new StringBuilder();

    while ((readCount = stream.read(buffer)) > -1) {
        builder.append(new String(buffer, 0, readCount));
    }

    return builder.toString();
}

private String buildPostData(HashMap<String,String> postData) throws UnsupportedEncodingException
{
    String data = "";

    for (Map.Entry<String,String> entry : postData.entrySet()) 
    {
        data += (URLEncoder.encode(entry.getKey(), "UTF-8") + "=" + URLEncoder.encode(entry.getValue(), "UTF-8") + "&");        
    }

    // Trim the last character (a trailing ampersand)
    int length = data.length();

    if (length > 0) {
        data = data.substring(0, (length - 1));
    }

    return data;
}

private String buildJSONError(String message, String name, String at)
{
    String error = "{\"errors\":[{\"message\":\"" + message + "\",\"name\":\"" + name + "\",\"at\":\"" + at + "\"}]}";

    return error;
}

private String callPost(String url, HashMap<String,String> postData) throws IOException
{
    // Set up the URL for the API call 
    URL apiUrl = new URL(url);

    // Build the post data
    String data = buildPostData(postData);

    // Call the API action
    HttpURLConnection conn;

    try {
        conn = (HttpURLConnection)apiUrl.openConnection();
    } catch (IOException e) {
        throw new IOException(buildJSONError("Failed to open a connection.", "CONNECTION_FAILURE", ""));
    }

    // Set connection parameters for posting data
    conn.setRequestMethod("POST");
    conn.setUseCaches(false);
    conn.setDoInput(true);
    conn.setDoOutput(true);

    // Write post data
    try {
        DataOutputStream wr = new DataOutputStream(conn.getOutputStream());
        wr.writeBytes(data);
        wr.flush();
        wr.close();
    } catch (IOException e) {
        throw new IOException(buildJSONError("Failed to post data in output stream (Connection OK?).", "POST_DATA_FAILURE", ""));           
    }

    // Read the response from the server                
    InputStream is;

    try {
        is = conn.getInputStream();
    } catch (IOException e) {
        InputStream errStr = conn.getErrorStream();

        if (errStr != null) 
        {
            String errResponse = readInputStream(errStr);
            throw new IOException(errResponse);
        } 
        else 
        {
            throw new IOException(buildJSONError("Failed to read error stream (Connection OK?).", "ERROR_STREAM_FAILURE", ""));
        }
    }

    // Read and return response from the server
    return readInputStream(is);
}

/* -------------------------------------
 * 
 * Synchronous API calls
 * 
   ------------------------------------- */

public APIResponse<CustomerAuthentication> authenticateCustomer(HashMap<String,String> postData)
{
    // Set the URL for this API call
    String apiURL = m_apiDomain + "/customer/authenticate" + m_apiContentType;

    Gson jsonConv = new Gson();

    String apiResponse = "";

    try 
    { 
        // Call the API action
        apiResponse = callPost(apiURL, postData);

        // Convert JSON response to the required object type
        CustomerAuthentication customerAuth = jsonConv.fromJson(apiResponse, CustomerAuthentication.class);

        // Build and return the API response object
        APIResponse<CustomerAuthentication> result = new APIResponse<CustomerAuthentication>(true, customerAuth, null);

        return result;
    } 
    catch (IOException e) 
    {
        // Build and return the API response object for a failure with error list
        APIErrorList errorList = jsonConv.fromJson(e.getMessage(), APIErrorList.class);

        APIResponse<CustomerAuthentication> result = new APIResponse<CustomerAuthentication>(false, null, errorList);

        return result;
    }
}

}

Answer 1

If you are getting an error it could be because you are overloading the authentication service (something which doesn’t happen if you do this one at a time) Perhaps it returning a error like 500, 503 or 504 which you could be ignoring and getting nothing you expect back, you return null http://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html

I would use less threads assuming you don’t have 1000 cpus, its possible having this many threads will be slower rather than more efficeint.

I would also check that your service is returning correctly every time and investigate why you get a null value.

If your service can only handle say 20 requests at once, you can try using a Semaphore as a last resort. This can be using to limit the numebr of concurrent requests.