Wiring up Model-View-Presenter UIs using Google Gin

We make heavy use of Google Web Toolkit in our dashboard systems: Analytics, management, configuration, and so on. In order to make these applications clean, testable, and modular, we apply the Model-View-Presenter pattern. However, MVP alone is not a particularly good pattern – views are tied to presenters by a controller class and have a circular dependency with their presenters. Changes to the system can require altering multiple moving parts.

This is where we’ve applied Google Gin as a way to create a more loosely coupled implementation of MVP. We still have to put up with the view/presenter circular dependency, but it is cleaner and more manageable now.

Here is how:

First, the GWT module initializes the dependency injection cascade by asking an injector we wrote for the controller class:

    private final AppGinjector injector = GWT.create(AppGinjector.class);
    public void onModuleLoad() {
        injector.getDashboardController();
    }

Read the rest of this entry »

Refactoring away Singletons with Google Guice

We’ve been engaged on a number of projects lately, such as integrating the and jsTestDriver on the front-end and rolling out Guice and ASM on the back-end.

One of the most enjoyable aspects of these projects has been refactoring the Singletons in our system: We see the typical implementation of Singletons as an anti-pattern that creates tightly bound systems with tangled dependencies.

Rather than showing our own code, I wrote a simple project to illustrate one way we get rid of Singletons. The Main function starts up a pair of Singletons, lets them do some work, and prints out the details:

   
        EventListenerService.get().initialize();
        EventGeneratorService.get().start();
        Thread.Sleep(100);
        for(Event e : EventListenerService.get().getRecordedEvents()) {
                System.out.println(e.info);
        }

Read the rest of this entry »

On Layering

While we only sell a single product – our Social Commerce suite of applications, internally, we’ve had to build a number of commercial-grade systems, some of which could be seen as products in and of themselves.

Product 1: First of all, in order to deliver components into remote websites at massive scale, we’ve had to build a sophisticated platform that can deliver content across the web, and integrate data from multiple websites while remaining cross-browser compliant. This platform is built on Java, AWS, CloudFront, Memcached, and other sophisticated “shoulders of giants” systems.

Product 2: Next, in order to deliver analytics over these components, which can be loaded thousands of times per second, we’ve had to develop an analytics platform based on Pig, Hive, Hadoop, S3, and Postgres. This is a layered system, providing multiple tiers of services. We’ve also had to develop our own link shortener.

Product 3: The most visible of our products is our suite of social applications. These build on top of the other two products to deliver trackable, scalable, personalized social content to users. This system, built using JavaScript, YUI, Freemarker, Java, Ehcache, Memcached, and Postgres, sits on a complex multi-tenanted graph database that tracks social graphs and provides deep personalization data.

By treating each part of the system as a standalone product, and building each one to the highest level of performance and API quality, we ensure that new products can be rapidly built by leveraging existing technology investments, which allows us to innovate at a business level – where, at the end of the day, it really counts.