ROSSEN STOYANCHEV SPRING FRAMEWORK DEVELOPER In the past year Netflix shared a story about upgrading their main gateway serving 83 million users from Servlet-stack Zuul 1 to an async and non-blocking Netty-based Zuul 2. The results were interesting and nuanced with some major benefits as well as some trade-offs. Can mere mortal web applications make this journey and moreover should they? The best way to explore the answer is through a specific use case. In this talk we'll take 5 common use cases in web application development and explore the impact of building on Servlet and Reactive web application stacks. For reactive programming we'll use RxJava and Reactor. For the web stack we'll pit Spring MVC vs Spring WebFlux (new in Spring Framework 5.0) allowing us to move easily between the Servlet and Reactive worlds and drawing a meaningful, apples-to-apples comparison. Spring knowledge is not required and not assumed for this session.

1. Rossen Stoyanchev
Servlet vs Reactive
5 use cases
stacks
in

2. reactive

3. asynchronous

4. Blocking
component

5. Blocking
component
Thread pool

6. Async component
?
concurrency strategy?

7. [elastic] thread pool
Async component
Thread pool

8. Async component
(non-blocking)
event loop

9. Async throughout (not an afterthought)
Code needs to evolve to adapt to non-blocking execution
Efficient use of resources, scale, resilience
Reactive brings flow control

10. How Java can help
Future: forget about it
CompletableFuture: good but single value only
Stream: for collection but no latency
Flow (Java 9): a.k.a. Reactive Streams

11. Reactive libraries
Reactor, RxJava
Reactive Streams interop with back pressure across
Functional composition of async logic
Intro talks: QCon, Devoxx, SpringOne

12. Spring Framework 5.0
Ready for JDK 9
Functional with Java 8 and Kotlin
Java EE 8 integration (Servlet 4.0, JPA 2.2, Validation 2.0, JSON Binding)
Reactive programming model

13. Servlet Stack
● Servlet container
● Servlet API
● Spring MVC
Reactive Stack
● Netty, Servlet 3.1+, Undertow
● Reactive Streams, Reactor
● Spring WebFlux

14. Reactive Spring
Reactive starters in Spring Boot 2.0
Spring WebFlux endpoints + reactive HTTP client
Reactive Spring Data Kay repositories
Spring Security
and more…

15. models
Execution
Stack

16. FilterFilter Controller
Spring
MVC
Servlet
container thread
Servlet API “thread-per-request” model

17. FilterFilter Controller
Spring
MVC
FilterFilter Controller
Spring
MVC
… do work on any thread, or wait for event …
AsyncContext.dispatch()
Servlet
container thread
Servlet
container thread
SERVLET 3.0 ASYNC …

18. Servlet stack
Thread pool powered, synchronous Servlet API
Async request handling is an option
Blocking network I/O

19. Spring
WebFlux
Reactive stack
Web
Filter
NO
BLOCKING
ANY
TIME
Controller
Reactive
Server
Adapter
Web
Filter
HTTP Server
Event Loop

20. Even loop
worker
FMonoEFluxDMonoCMonoB
Controller
Spring
WebFlux
WebFilter
1. Composition of async logic
MonoA
A B C
EF D
2. Actual processing

21. Event loop driven
Code must be non-blocking
Non-blocking I/O with back pressure
Reactive stack

22. repository
Use Case #1
dataReactive

23. Demo

24. HTTP GET with reactive data repository
Designed to work on both Spring MVC and Spring WebFlux
Simply return reactive type (Flux, Observable) from @Controller

25. Flux<T>:
finite collection or infinite stream?

26. Use media type to decide

27. “application/json”
finite collection (JSON array)

28. No back pressure:
Flux#collectToList
(request all + buffer)

29. “text/event-stream”,
“application/stream+json”
infinite stream

30. Back pressure:
request(n),
write, flush,
repeat

31. Use Case #2
back pressure
with
Response stream

32. HTTP GET with streaming response
Simply return reactive type (Flux, Observable) from @Controller
Back pressure on Spring MVC and WebFlux

33. Servlet
Container
Servlet
API
Spring
MVC
Flux
Back pressure against blocking OutputStream
Controller Flux
Data
Repository
onNext(T)onNext(T)
Blocking write on
MvcAsync thread pool
request(n)request(1)
SERVLET STACK …

34. Servlet 3.1 non-blocking I/O ?
Unlike Servlet 3.0 async, Servlet 3.1 non-blocking is hard to retrofit
Requires deeper change
Mutually exclusive with rest of the Servlet API

35. HTTP
Server
Server
Adapter
Flux
Spring
WebFlux
Flux
Response streaming on reactive stack
Controller Flux
Data
Repo
onNext(T)onNext(T)onNext(T)
Non-blocking write
back pressure
to socket
request(n)request(n)request(n)

36. Demo

37. Use Case #3
remote service
orchestration
Reactive

38. Demo

39. Orchestrate non-blocking, nested remote service calls with ease
Similar to reactive data access
Spring MVC and Spring WebFlux
Reactive WebClient

42. Use Case #4
request inputReactive

43. Back pressure to socket
No reading until reactive demand signalled from upstream
Non-blocking
Reactive stack only territory !

44. HTTP POST with data
@RequestBody argument with reactive type (Mono, Single)
Reactive type is not required

45. Use Case #5
Data Ingestion
back pressure
with

46. Media type indicates infinite stream is expected
Non-blocking streaming + back pressure
HTTP POST with stream of data

47. HTTP
Server
Server
Adapter
Flux
Spring
WebFlux
Flux
Data ingestion on reactive stack
Controller
onNext(T)onNext(T)
Non-blocking read
back pressure
from socket
request(n)request(n)

48. Servlet stack summary
Reactive data repository
Streaming to the response with back pressure
Reactive orchestration of remote services
Reactive request input
Data ingestion with back pressure

49. Reactive stack summary
Reactive data repository
Streaming to the response with back pressure
Reactive orchestration of remote services
Reactive request input
Data ingestion with back pressure

50. Q & A
@rstoya05