Take the 2-minute tour ×
Programmers Stack Exchange is a question and answer site for professional programmers interested in conceptual questions about software development. It's 100% free, no registration required.


  • One data input feed (call this Source)
  • Multiple objects (call these Layer1) read this Source, and analyze the feed. This analysis is stateful.
  • Layer1 objects have configuration parameters. For example, new FooLayer1(10, 20) will do different analysis than new FooLayer1(20, 20) will do different analysis than new FooLayer1(20, 30)
    • The Layer1 analysis is done asynchronously.
  • Still other objects (call these Layer2) readonly the output of the Layer1 objects and process those results. This analysis is also stateful.
  • Guice object creation hierarchy
  • Needed Layer1 objects are accessed often enough that it doesn't matter if they are created lazily, as long as they remain in memory after they're created the first time.
  • The needs of Layer2 objects are hardcoded; no good way to know in advance which Layer1 objects will be needed.

I want to be able to reuse the work of the Layer1 objects. For example, lets say I have:

  • AlphaLayer2 object that reads the output of FooLayer1 and BarLayer1
  • BetaLayer2 reads the output of BarLayer1 and BazLayer1
  • QuuxLayer1 is implemented, but I don't want my server to compute the value of QuuxLayer1 unless there is a Layer2 object that needs it.

Currently, I have a factory that all Layer2 objects have access to, that only creates the appropriate Layer1 object as needed, and stores all created Layer1 objects in a Map. But, since Layer1 objects also have configuration parameters, I need to make sure that I'm storing the Layer1 objects in a good data structure. But the Layer1 functionality needs to be flexible, and I figure as I create more kinds of Layer1 objects, this kind of a Map could get really complicated. I am asking this question because, while it's fine now, as the complexity grows, maintaining the data structure in this way seems to become really difficult.

Trying to do some sort of nested Map data structure should get me O(1) access times on finding these Layer1 objects, but will make the caching super complicated. Alternatively, I could keep them in a much flatter structure, and iterate over the list of all the Layer1 objects and see if they have the right parameters. If none found, create the new one. As long as Layer2 objects keep the reference, this should only happen once. Yet another option would be to create a Map<Class, Set<Layer1>> so that it only needs to iterate through the created Layer1 objects of the correct class.

Has anyone seen this kind of scenario before? Which of the options I've mentioned is preferred? Or, is there something I haven't mentioned yet? Also, is there any way to use Guice to create Layer1 objects in a way where I can get reuse them (which I need to do, because they're stateful)

share|improve this question
Do you have scalability constraints? Have you considered making the computation distributed? Have you looked at storm? It uses reliable streaming in a predefined topology of computation layers that would allow you to accomplish what you described at scale. –  itaifrenkel May 11 '13 at 6:15
@itaifrenkel very cool. I have a lot of code written already, I'm not sure it's best to throw everything out right now. It's already very asynchronous, I'm using MBassador –  durron597 May 11 '13 at 19:18
add comment

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.