beyondtheflux-150927075557-lva1-app6892.pdf

1. Thought about beyond Flux Layering an application for scalability Tetsuharu OHZEKI Github:saneyuki
2. CAUTION • This is still a note of my thought. • Some points are unclear and opaque. • I believe that we should practice a way which might not get a glory but never lose for building an application.
3. Keypoints of Flux “pattern” • Unidirectional data flow + observer pattern + command pattern • It’s just a Model-View-Intent. • http://staltz.com/mvi-freaklies/ • WPF’s “Command” is similar terminology. • One of design patterns than an architecture.
4. Escape from “Flux” enclosure! Flux is just a pattern!
5. Separate a “Responsibility” • We should separate a “Responsibility” rather than “Concern”. • The most important point is “cross over the boundary of responsibility”. • In other word, this “responsibility” is a “context”.
6. Separate an internal “layer” • An internal design is layered naturally if you separate the boundary of a responsibility. • Typical cases: OS’ layering architecture, OSI reference model. • Hardware abstraction, low level API, providing a process lifecycle and a higher level API, running an actual application domain based on APIs, present a computation result, and etc.
7. Example: Windows 7 Architecture Mark E. Russinovich, David A. Solomon, Alex Ionescu（2012）『インサイドWINDOWS 第6版 上』株式会社クイープ訳, Microsoft Press, p53.
8. Some typical layerings
9. User input (mouse event) command (e.g. Flux Action) domain model view state model View presentation Update UI by some user action
10. User input (mouse event) command (e.g. Flux Action) Update view state model Update view presentation Update only UI by some user action
11. User input (mouse event) command (e.g. Flux Action) domain model REST API abstraction HTTP abstraction (and TCP abstraction… ?) Dispatch a HTTP request by user action
12. HTTP abstraction (and TCP abstraction… ?) REST API abstraction Update domain model Update view model Update view presentation Receive a HTTP response & Display it
13. Each “layer” is a data transform layer • A data flow in application is a stacked data transformations which is a result of each layers. • As a simplified concept, their basis is a result of many map/flatmap computations.
14. The benefit of Reactive Extensions • Unify an asynchronous interface. • Make each layer boundaries to Observable<T>. • We can compute Observable<T> with using their operators. • Code a data flow graph which are produced from asynchronous events.
15. Layering example Driver Adapter Domain Domain Model Gateway HTTP Socket Repository DB Driver View Model View Command (Intent)
16. Is this a Domain Driven Development? • My interpretation is that domain driven development is one of practical system. • I feel there are mixed into some system-specific practices and generic practices. • This session focuses into a generic thought about an architecture • e.g. layering.
17. This vision is too rich! • This is a high level abstract concept. • Rightfully, some hardware/environment/performance limitations limits a materialization of this concept. • Software engineering is just filling this gap, and the flower of it.
18. Where is the framework? • This does not suppose any framework libraries. • It’s inevitable that All software codebase/repository make some style/framework in them. • The difference is whether cutting out internal materials from it.
19. dependency relation (1) • In Clean Architecture context, domain model layer should not depend on presentation (driver) layer. • I agree it as the basic principle. • But it’s hard to apply it strictly…
20. dependency relation (2) • There need a “connection layer” which connects HAL and specific domain layer. • Some runtime supports like dependency injection would cover and help this connective operation. • So it would be difficult that all relationships among layers are a unidirectional dependency perfectly. • We cannot defined a simple rules :(
21. dependency relation (3) • In a general usecase, its rule should be confirmed that “domain layer don’t touch a driver layer directly”. • However, of course, we would need to accept some dirty pit holes to resolve performance sensitiveness.