Recursive Architecture

Transcript

1. Recursive Architecture
2. Josh Comer Math Dad Architect at LiveOps Cloud Tea Enthusiast
3. What is Recursion
4. Recursions 101 Definition: see recursion Real Definition: the repeated application of a procedure or definition Two mirrors, fractals, etc.
5. CS Recursion Everyone’s favourite topic in school Remember recursion == iteration Fib(n) = Fib(n-1) + Fib(n-2), Fib(0)=0, Fib(1)=1 Fib is interesting case as naïve recursion leads to O(phi^n) while naïve iteration is O(n), though easy to fix.
6. Quicksort Divide and Conquer (defn qsort3 [[pivot :as coll]] (when pivot (lazy-cat (qsort (filter #(< % pivot) coll)) (filter #{pivot} coll) (qsort (filter #(> % pivot) coll)))))
7. So what’s the big deal Recursion is pure Recursion is clean Recursion is terse The barrier to solving a problem recursively usually means the solution has been thought through
8. Conway’s Law organizations which design systems ... are constrained to produce designs which are copies of the communication structures of these organizations — M. Conway
9. Dev-isms
10. Josh Comer Senior Distinguished Software Engineering Evangelist of the Architecture Variety
11. BDD
12. Blog Driven Development Someone who is greatly influenced by reading a single blog post… daily ...and then changes all the code
13. SODD
14. Stack Overflow Driven Development Someone who codes exclusively using Stack Overflow answers. I’m a SODD with bash ☺️
15. Heisenbug
16. Heisenbug A defect which only happens when not observing (debugging) the software
17. return;
18. Architecture
19. Why do we use patterns? Patterns help us communicate Humans are really good at patterns
20. But… Patterns all start to fall apart if everyone isn’t on the same page. Not everyone has the GoF book memorized
21. Concurrency The next frontier in programing: Vertical Scaling won’t work anymore But concurrent programming is hard for even the best programmers Abstract away the concurrency!!! - Joe Armstrong
22. OO Architecture Design from the abstraction towards the problem Good OO design requires a lot of upfront effort Distracts from the problem Harder to change (we don’t always get it right the first time)
23. Recursive Architecture Conway’s Law We are, seemingly, naturally good at reasoning about systems in the large Distributed Systems are becoming the new norm
24. Where is the recursive part? We can design our services using the same primitives as our platform The inside mirrors the outside
25. Let’s Design Something
26. <Insert cliché example>
27. Temporally Sorted Activity Planner ⏰
28. TSAP We need to make something that allows us to store tasks in a list (sorted temporally). In addition to the task list, we want to be able to schedule actions at particular times.
29. TSAP
30. TSAP API
31. TSAP API Load Balancer
32. TSAP API Load Balancer DB
33. TSAP API Load Balancer DB Scheduler
34. TSAP API Load Balancer DB Scheduler Notifier
35. TSAP API Load Balancer DB Scheduler Notifier(s) Queue
36. Scheduler Needs to be: Receive scheduling requests Store pending tasks Send triggered tasks to the notifier
37. Scheduler
38. Scheduler RPC
39. Scheduler RPC Ingress Server
40. Scheduler RPC Ingress Server DAL
41. Scheduler RPC Ingress Server DAL Journal
42. Scheduler RPC Ingress Server DAL Journal Egress Server
43. Scheduler RPC Ingress Server DAL Journal Egress Server Queue
44. Building Blocks There are three main building blocks: API Servers Processes Pub-Sub
45. Building Blocks There are three main building blocks: API Servers -> Gen-Server Processes -> Actor Pub-Sub -> Gen-Event Supervisors Gen-FSM It’s the Erlang OTP Behaviours
46. API Servers are key AXD301 (one of the biggest Erlang code bases) uses gen- servers for 63% of its behaviours APIs allow us to hide execution particulars (think façade pattern)
47. “You know nothing API Snow” An API can hide concurrency (it acts as a load balancer) An API can make asynchronous calls synchronous An API can provide a translation layer An API can handle errors An API can manage resiliency
48. Isn’t this an Interface? Not quite The API is an active piece of code that contains control logic For example gen-server has fns for: Sync Call Async Call Timeout Shutdown and Startup
49. Can we take this further? API Load Balancer DB Scheduler Notifier(s) Queue
50. Can we take this further?
51. So what is a service?
52. Mic Drop
53. Microservices Let’s call them nanoservices Even smaller than microservices Responsible for an extremely small portion of logic Built to be deployed in a shared resource environment with other nanoservices
54. Is this even a thing? Actors! Actors are back Erlang is making a comeback (check out Elixir) Akka or Quasar for Java Akka.net for .NET
55. Fin