1. Async Microservices
with Finagle
Vladimir Kostyukov
http://vkostyukov.ru

2. Asynchronous
Protocol-Agnostic
Full-Stack RPC
2

3. Adopters
3

4. Built atop of Netty
(#1 Non-Blocking I/O for JVM)
4

5. Finagle -
Netty in a functional setting
5

6. Building Blocks:
Services + Futures + Filters
6

7. trait Service[-Req, +Rep] extends (Req => Future[Rep])
7

8. Service-Oriented
rather then
Stream-Oriented
8

9. // TODO: Don't forget to remove it in production.
object GetDeveloperPromotionDate extends Service[Developer, Date] {
def apply(req: Developer) = Future.never
}
val date: Future[Date] = GetDeveloperPromotionDate(Developer.Myself)
9

10. 1 val respond = new Service[HttpRequest, HttpResponse] {
2 def apply(req: HttpRequest) = {
3 val rep = new DefaultHttpResponse(HttpVersion.HTTP_1_1,
4 HttpResponseStatus.OK)
5 rep.setContentString(req.getUri())
6 Future.value(rep)
7 }
8 }
9
10 Await.ready(Http.serve(":8080", respond)
10

11. Servers and Clients
talk to each other via services
11

12. 1 val to = Http.newService("google.com")
2
3 val proxy = new Service[HttpRequest, HttpResponse] {
4 def apply(req: HttpRequest) = to(req)
5 }
6
7 Await.ready(Http.serve(":8080", proxy)
12

13. Are there high-order services?
13

14. 14
Client
Service
Server
Filter Filter Filter Service

15. // ReqIn -> Service(ReqOut -> RepIn) -> RepOut
trait Filter[-ReqIn, +RepOut, +ReqOut, -RepIn]
extends ((ReqIn, Service[ReqOut, RepIn]) => Future[RepOut])
15

16. // A filter that does nothing.
def identity[Req, Rep] = new Filter[Req, Rep, Req, Rep] {
def apply(req: Req, service: Service[Req, Rep]) = service(req)
}
val auth = identity[HttpRequest, HttpResponse]
16

17. Filters are type-safe!
17

18. A Type-Safe Input Filter
1 object TurnAIntoB extends Filter[A, Rep, B, Rep] {
2 def apply(req: A, service: Service[B, Rep]): Future[Rep] =
3 service(req.toB)
4 }
5
6 val serviceOfB: Service[B, Rep] = ...
7 val serviceOfA: Service[A, Rep] = TurnAIntoB andThen serviceOfA
18

19. A Type-Safe Output Filter
1 object TurnAIntoB extends Filter[Req, B, Req, A] {
2 def apply(req: Req, service: Service[Req, A]): Future[B] =
3 service(req) map { a => a.toB }
4 }
5
6 val serviceOfA: Service[Req, A] = ...
7 val serviceOfB: Service[Req, B] = TurnAIntoB andThen serviceOfA
19

20. Case Study: TimeOut Filter
1 def timeout[Req, Rep](timeout: Duration)(implicit timer: Timer) =
2 new SimpleFilter[Req, Rep] {
3 def apply(req: Req, service: Service[Req, Rep]) =
4 service(req).within(timer, timeout)
5 }
6
7 val handleExceptions = new SimpleFilter[Req, Rep] {
8 def apply(req: Req, service: Service[Req, Rep]) =
9 service(req) handle {
10 case t: TimeOutException => Future.value(defaultRep)
11 }
12 }
13
14 val respond = new Service[Req, Rep] { ... }
15 // We guarantee 3 seconds response time.
16 val backend = handleExceptions andThen
17 timeout(3.seconds) andThen
18 respond
20

21. Case Study: OAuth2Filter
1 class OAuth2Filter[U](dataHandler: DataHandler[U])
2 extends Filter[HttpRequest, HttpResponse, OAuth2Request[U], HttpResponse]
3 with OAuth2 with OAuthErrorHandler {
4
5 def handleError(e: OAuthError) = e.toHttpResponse
6 def apply(req: HttpRequest, service: Service[OAuth2Request[U], HttpResponse]) =
7 authorize(req, dataHandler) flatMap { authInfo =>
8 service(OAuth2Request(authInfo, req))
9 } handle {
10 case e: OAuthError => handleError(e)
11 }
12 }
13
14 val authorize = new OAuth2Filter(...)
15 val respond: Service[OAuth2Request[U], HttpResponse] = ???
16 val backend: Service[HttpRequest, HttpResponse] = authorize andThen respond
21

22. Future is a monad!
22

23. Abstraction: Fibonacci Calculator
1 trait FibonacciCalculator {
2 val Zero = BigInt(0)
3 val One = BigInt(1)
4 val Two = BigInt(2)
5
6 def calculate(n: BigInt): Future[BigInt]
7 }
23

24. Sequential Composition: map & flatMap
1 trait Future[+A] {
2 def flatMap[B](f: A => Future[B]): Future[B]
3 def map[B](f: A => B): Future[B]
4 }
24

25. FlatMap Power
1 val ab: Service[A, B] = ???
2 val bc: Service[B, C] = ???
3 val cd: Service[C, D] = ???
4
5 val req: A = ???
6 val rep: Future[D] = ab(req) flatMap bc flatMap cd
25

26. Sequential Composition: map & flatMap
1 object FlatMapFibonacciCalculator extends FibonacciCalculator {
2 def calculate(n: BigInt): Future[BigInt] =
3 if (n == Zero || n == One) Future.value(n)
4 else calculate(n - One) flatMap { a =>
5 calculate(n - Two) flatMap { b => Future.value(a + b) }
6 }
7 }
8
9 object MapFibonacciCalculator extends FibonacciCalculator {
10 def calculate(n: BigInt): Future[BigInt] =
11 if (n == Zero || n == One) Future.value(n)
12 else calculate(n - One) map { a =>
13 calculate(n - Two) map { b => a + b }
14 }
15 }
26

27. Sequential Composition: for-comprehension
1 object ForFibonacciCalculator extends FibonacciCalculator {
2 def calculate(n: BigInt): Future[BigInt] =
3 if (n == Zero || n == One) Future.value(n)
4 else for {
5 a <- calculate(n - One)
6 b <- calculate(n - Two)
7 } yield a + b
8 }
27

28. Concurrent Composition: collect & select
1 object Future {
2 def collect[A](fs: Seq[Future[A]]): Future[Seq[A]]
3 def select[A](fs: Seq[Future[A]]): Future[A]
4 }
28

29. Concurrent Composition: collect
1 class FanoutFibonacciCalculator(
2 left: FibonacciCalculator,
3 right: FibonacciCalculator) extends FibonacciCalculator {
4
5 def calculate(n: BigInt): Future[BigInt] =
6 if (n == Zero) || n == One) Future.value(n)
7 else {
8 val seq = Seq(left.calculate(n - One), right.calculate(n - Two))
9 Future.collect(seq) map { _.sum }
10 }
11 }
29

30. Concurrent Composition: select
1 class SelectFibonacciCalculator(seq: Seq[FibonacciCalculator])
2 extends FibonacciCalculator {
3
4 def calculate(n: BigInt): Future[BigInt] = {
5 val futures: Seq[Future[BigInt]] = seq map { _.calculate(n) }
6 Future.select(futures)
7 }
8 }
30

31. Futures should be
chained asynchronously
31

32. Do Not Await
// Asynchronous operation.
val f = service(req)
// Blocking operation.
Await.result(f)
32

33. Involving Side-Effects
// Asynchronous operation.
val f = service(req)
!
f onSuccess { rep => println("Got the value: " + rep) }
onFailure { t => t.printStackTrace() }
33

34. Finagle is a non-blocking glue for
services that implement
different protocols
34

35. Finagle is extremely good at
gluing things that work
with different speed
35

36. References
§ http://twitter.github.io/finagle/
!
§ http://vkostyukov.ru/posts/finagle-your-fibonacci-calculation/
!
§ https://github.com/finagle/finch
!
§ https://github.com/finagle/finagle-oauth2
36

37. Stay Finagled!
!
And drop your feedbacks to
@vkostyukov
37