Sunday, July 28, 2013

Java Memory Troubles? Tweak the GC

tl;dr version

Memory issues? Try -XX:+UseSerialGC -Xss8m in your Run Configuration or as command-line parameters.

The Problem

I'm guessing that, like myself, you have found yourself struggling with the following, two memory errors. I will share what I found as solutions to them.

java.lang.OutOfMemoryError: Java heap space

So, you've taken great pains to use only primitive data types, switching from using an ArrayList<ArrayList<Integer>> for your data type to  Integer[][] even though it's not nearly as easy to work with. You discovered, though, that this just delayed running out of heap space. I tried a solution stated in many other posts on the web and that did seem to help but I can tell you that it didn't fix the problem. Rather, it allowed me to get to the next error.


I encountered this error for the first time when working on a project with a lot of recursion. This was likely due to what data I was inadvertently putting onto the stack during recursion. However, I recently worked on another project where I couldn't optimize any further the data pushed onto the stack during recursion. Moral of the story is that you should take the time to profile your code and optimize the data which may end up on the stack during recursion. However, after doing your due diligence, you still may need help, though not as much as you might think.

A Likely Solution

After a lot of research, I found the following two posts which helped tremendously. First, the solution and then the links. Add the following to your command line when executing your class or add it to your Run Configuration in Eclipse:
-XX:+UseSerialGC -Xss8m


This first switch is the one that intelligent developers worry will be overused. The default for the stack is 512KB. Suggestions have been as high as 1024m for the number component of this parameter. That's 1GB or RAM! If you need that much RAM for your stack then you might be doing it wrong. Something to note is that I had used this on the highly recursive project I mentioned first with 16m as my value. That got things working. Then I went back and got rid of a bunch of Strings in my code which I'd been using for debugging and I no longer needed this flag at all. Moral of the story, don't use it unless you have to so that you are forced to optimize your code.


It turns out that the java.lang.OutOfMemoryError: Java heap space error isn't always what it appears to be. Sure, I was running out of heap space but increasing heap space just allowed bad behavior to continue. A flag which is commonly recommended to deal with this error is -Xmx1024m. Again, that's 1GB of RAM. Now, the heap is where your application runs and by default it is given 64MB to work with. So, you could have a valid argument that one day you'll create an application which needs more heap memory. Just not today. It turns out, and this is keyyou just might not need more memory at all!
See, the problem is that Java's garbage collection, while very useful in letting us not worry about destructors and generally functional, can be lazy. So, this is a hint to the GC on how to run. By using this flag, I made it completely unnecessary to use the -Xmx flag. Further, this flag meant that I didn't have to have the -Xss flag set very high, either. Sure, I'm not writing a 4KB demo but I'm pretty happy to have everything working in less than 9MB.

Final Words

I hope you find these flags useful. The key thought, though, is that before you put any of the flags on this page to use, inspect your code with careful attention to the data types used. You may just find that you can slim down a bit and use far less memory in doing so. Remember, one day you might be writing code for a mobile or embedded device smaller than a bracelet, i.e. FitBit, and memory costs money and takes up space.

Tuesday, July 9, 2013

Strategy Pattern using Onion Architecture Templified

The following article was drafted 2013/07/09 shortly after a presentation I had made at the Orlando .NET Users Group. Unfortunately, I neglected to hit the Post button and left it to collect dust without being published until now. Apologies for anyone who was looking forward to this article.

Download demo code

[On 2013/07/09] I presented the strategy pattern at the Orlando .NET Users Group monthly meeting. The gist of it is that if you find yourself starting to write a switch statement or a cascading series of if/else statements, you need to stop for a moment and ask this question: "Will I be having to revisit this someday? If so, how often?" Your answer will dictate if you should proceed merrily along or implement the strategy pattern.

I had the opportunity to ask myself this very question recently when I was required to write a set of web scrapers to get price and availability information off of various retailers' product pages. Some retailers implement the standard for encoding this product data and others use a proprietary encoding. When my process gets the work item, it includes the URL and the encoding type so I started writing the if/else statement to handle this. Then I re-read the spec and noticed that other schemas were coming down the pike so I started writing a switch statement to handle the ones listed. It was at that moment when I realized that I was introducing a maintenance headache and knew the strategy pattern was the solution.

So, why did I decide that? After all, the strategy pattern doesn't prevent the need for maintenance. What it does very well, though, is moves the maintenance out of the business or (shudder) UI logic, the logic which isn't likely to change often, and promotes it to its own set of classes which actually do the work. The business logic simply asks a context object for whichever class will fulfill the need of the piece of work being handled.

Let me demonstrate using a trivial example for simplicity. I have a web page for doing product searches. On that page, I allow the user to decide if they want to use Bing or Google. When they hit the Search button, the PerformSearch action is called.

private readonly IShoppingSearchAPI _shoppingSearch;

public ActionResult PerformSearch(QueryModel queryModel)
    _shoppingSearch.Context = queryModel.SearchEngine;
    var searchResults = _shoppingSearch.Search(queryModel.Query, 10);
    var content = "";
    content += "";

    foreach (var product in searchResults)
        var inventoryData = product.InventoriesData != null ? product.InventoriesData.First() : new InventoryData();
        var buyUrl = product.Link;
        content += string.Format("", 
            product.Brand, product.Title, product.Description, inventoryData.Price, inventoryData.Shipping, inventoryData.Tax, buyUrl);

    content += "...creates a table...";

    return Content(content, "text/html");

Key takeaways here:
  • The code here knows nothing about how many search engines there are to choose from.
  • The code doesn't make any decisions about the search engines.
  • The code does not need to be maintained when a new search engine is added.
Now, I want to make it clear that this is a contrived example. Only two choices exist and I doubt you'd be adding new search engines to a web page for the user to pick at runtime. However, this clearly demonstrates that such runtime choices needn't be handled in the UI or business logic. Rather, it can be pushed to the outside ring of the Onion Architecture so that maintenance can be managed much more easily.

So, if the code isn't in the UI or business "layers" (Core in my unlayered architecture, name TBD but currently still Onion Architecture Templified) then where does it live? It lives where all code expecting to change often lives, in Infrastructure with an interface in Core. Here's the interface:

public interface IShoppingSearchAPI
    SearchEngine Context { get; set; }
    Image Logo { get; set; }
    IEnumerable<Product> Search(string query, int maxResults);

And here is an implementation for Google:

public class GoogleShoppingSearchAPI : IShoppingSearchAPI
    public SearchEngine Context { get; set; }

    private Image _logo;
    public Image Logo
        get { return _logo ?? (_logo = Image.FromFile(@"%HOMEDRIVE%%HOMEPATH%\Documents\Visual Studio 2012\Projects\CodeCampDemo\Infrastructure\Services\Google\logo4w.png")); } 
        set { _logo = value; }

    public IEnumerable<Product> Search(string query, int maxResults)
        var products = new List<Product>();

        var service = new ShoppingService { Key = "AIzaSyBcWCofOByUzp-EekjUCMxa30h-D9U95eE" };
        var request = service.Products.List("public");
        request.Country = "us";
        request.Q = query;

        var startIndex = 1;
        while (products.Count < maxResults)
            request.StartIndex = startIndex;
            var response = request.Fetch();
            if (response.CurrentItemCount == 0)

            foreach (var item in response.Items)
                var images = item.ProductValue.Images;
                var image = string.Empty;
                if (images != null)
                    image = images[0].Link;
                var product = new Product
                    Id = item.Id,
                    Brand = item.ProductValue.Brand,
                    Author = item.ProductValue.Author.Name,
                    Title = item.ProductValue.Title,
                    Description = item.ProductValue.Description,
                    Image = image,
                    Link = item.ProductValue.Link,
                    Gtins = item.ProductValue.Gtins

                if (item.ProductValue.Inventories != null)
                    product.InventoriesData = new List<InventoryData>();
                    foreach (var inventoryData in item.ProductValue.Inventories)
                        var inventory = new InventoryData
                            Price = inventoryData.Price,
                            Tax = inventoryData.Tax,
                            Shipping = inventoryData.Shipping,
                            Availability = inventoryData.Availability


                if (products.Count == maxResults)

        return products;

Finally, here is a typical context implementation, the place where your switch statement lives:

public class SearchEngineContext : IShoppingSearchAPI
    public SearchEngine Context { get; set; }

    public Image Logo { 
        get { throw new NotImplementedException(); }
        set { throw new NotImplementedException(); }

    public IEnumerable<Product> Search(string query, int maxResults)
        IShoppingSearchAPI searchEngineInstance;

            case SearchEngine.Bing:
                searchEngineInstance = new BingShoppingSearchAPI();
            case SearchEngine.Google:
                searchEngineInstance = new GoogleShoppingSearchAPI();

        return searchEngineInstance.Search(query, maxResults);

This implementation of the context likely differs from the examples you'll find on the 'net due to the fact that it implements the same interface as the classes it serves. This simplifies execution by making it unnecessary to ask for an instance before executing the action, doing so immediately instead. I use this improvement in the sample code to leverage .NET reflection features so that I don't even have to maintain the switch statement. Be sure to download the sample code to see how that was done.

So, the next time you find yourself creating a lengthy switch statement or a list of cascading if/else statements, ask yourself if you'll ever have to modify that code. If the answer is yes, then consider spending a little time implementing the strategy pattern and, if you want maximum flexibility, take a look at the demo code for how to do so in such a way that you could just add new classes to your infrastructure code and nothing more for new, related functionality to automagically appear.

Note: Technically, the demo also requires an Enum to be updated in Core but it would be a simple matter to use a string rather than an Enum to enable the same functionality.