I was studying NI's CLA sample examples for the ATM problem. I took NI's Actor Framework training and have started to use it in production code I'm writing. It's pretty obvious to me that the ATM example is reinventing much of what the actor framework does and I wonder if the graders are fine with us just leveraging AF to do much of the work that the ATM example is laboriously re-creating. E.g. Would it be fine to have Main launch a root actor which then launches nested actors? I would not have the user event that represents a global stop the way the solution modules do, instead I would simply send stop core to the root actor and expect that to stop all nested actors. Then I would define methods and corresponding messages for the actions (what are defined as Action Enums in the solution). Is that an acceptable approach that would score well? For errors, I think I could just use the Handle Error (I believe that's the AF name for it) override in the root actor such that it gets errors from all of it's nested children. That would eliminate the need to define a specific error handling module and send errors to said module (the way it's done in the example solutions). Am I right? I would need to have a way for all actors to access other actors enquers. The easiest way would be to have a global variable for each modules enqueuer (or an action engine) but I'm wondering if I would be scored poorly for such a global based approach? I believe a more correct method is to use the actor tree approach such that actors communicate to the root actor. I think the controller might for instance send a message to parent (root actor) to "send this (Database specific) Message to the Database module"? I have limited experience using multiple non-root Actors so I'm not sure if that's the right approach and I'm not sure if that's worth the trouble in terms of points on the exam (like would I score significantly better by doing that rather than just using the global(s) to access modules enqueuers)? Finally, I have a rather broad question and I wonder if someone can point me to a white paper or something. I don't entirely feel comfortable with the Asyncronous approach to programming. For example, my imperative programmer mind thinks Widthdrawl should be defined as: "check balance, if > withdraw amount then balance = balance - withdraw amount, dispense cash, display new balance". However, Actor Programming does not allow for straightforward function calls such as "get balance", instead I have to send a message to query the balance which then (maybe) results in a message to me containing account balance. At that point I'm no longer in the withdraw function. So instead I have to distribute around the logic for how to do a withdraw. In looking at the ATM example it is not at all clear to me how the solution does this. I see a "Controller:Withdraw Init" state which has a "Withdrawal" enum hovering in mid air (why?) and then Widthdraw Approved and Withdraw Failed states. So, I'm asking for a pointer to knowledge about how to design these Actor/Messaged based architectures such that functions like Withdrawing can be done in a distributed but deterministic way. Perhaps a fully working implementation of the ATM example would help if there is such a thing.

Certification

Hello Everyone,

I have made attempt with CLA ATM sample exam. Kindly review and let us know your comments.

Thanks

Download All

Message 11 of 11

(771 Views)