Typed Services Using Finch

Finch is an open source HTTP library built on top of Finagle, the RPC framework that powers Twitter's infrastructure. Finch is a great candidate to use when building services, and compares favourably to other popular frameworks in languages such as Ruby, Go, JavaScript, Elixir, Clojure and Haskell. This talk outlines the types of problems faced when building small services, and how using a good type system can help. I introduce & outline Finch, highlighting how it addresses these concerns. Throughout, a concrete example of an API built for a startup using Finch is used.

Transcript

1. TYPED SERVICES USING FINCH Tom Adams, @tomjadams YOW West 2016
2. 1 SERVICES
3. SERVICE • We care about things like • HTTP primitives • Request/response encode/decode • Transport protocols • Talking to downstream services • Local data storage • But not these • Asset packaging • View rendering • JavaScript, SASS, LESS, etc.
4. LANDSCAPE • Go: gokit, Kite • Elixir: Phoenix • Javascript: Node.js (TypeScript) • Clojure: Caribou, Liberator, Rook • Ruby: Rails, Sinatra, Grape, Lotus • Erlang: Leptus, Yaws • Haskell: Snap, rest, servant, Hapstack, Yesod • Java: Play, Spring, Jersey, Spark, RESTEasy, (Dropwizard) • Swift: Swifton, Vapor
5. WHAT’S IN A HTTP FRAMEWORK?
6. WE NEED 1. Routing • path/headers/methods/etc. to a function 2. Req → Resp
7. SERVICE PRIMITIVES • Routing is a function • r :: URI → a • An “action” is a function • a :: Req → Resp • A “controller” is (scoped) a collection of actions • c :: List a • A “service” is a collection of controllers • s :: List c
8. BUT WAIT, THERE’S MORE • HTTP primitives • Datastore • Metrics • Logging • JSON codec • Databinding • Configuration • Environments • HTTP clients • Failure isolation • Async primitives • Monitoring • Service discovery • Debugging/ tracing • Caching • Messaging • Deployment • Testing • Live code reload • …
9. 2 SCALA Let’s talk about Scala https://github.com/OlegIlyenko/scala-icon
10. WHY SCALA? • JVM support - “Better Java” • Fast, scalable • Deployment & runtime behaviour well understood • Library & tool support (distributed heap, debugging, etc.) • Decent (not great) static type system • Succinct - closures, type classes, type aliases, type inference, no semi-colons • Features - immutability, equational reasoning, functions, case classes, implicits, packages, mixins, currying/partial application, etc. • Standard library - option, either, future, etc. • Cool stuff! scalaz, actors, higher-kinds, etc.
11. WELL USED • Twitter, Pinterest, Airbnb, SoundCloud, Uber, Strava, Gilt, LinkedIn, Amazon, Tumblr, Foursquare, Box, Gigya, Simple, Localytics, LivingSocial, eHarmony, Yammer, Firebase, Disqus, Asana, Hootsuite, PagerDuty, Rdio, Mesosphere • Apple, Novell, The Guardian, Sony, BSkyB, AOL, Xerox, Siemens, VMware • REA, Seek, Skedulo, CBA, Atlassian, Fairfax, RedBalloon, Canva*, Oomph* Source: Quora, AngelList, scala-lang.org, reddit, LinkedIn, Finagle Adopters
12. WHY FP? • (Static) Types, and • Immutability, and • Composition, gives rise to • Equational reasoning, and • Certainty, and • Reliability
13. FRAMEWORK OPTIONS • Karyon (Netflix) • Play (Typesafe/Lightbend) • Unfiltered (OSS) • Dropwizard (Yammer) • Spray (Typesafe/Lightbend) • Finagle (Twitter) / Finatra (OSS) / Finch (OSS) • Akka, Lagom (Typesafe/ Lightbend) • Colossus (Tumblr) • Chaos (Mesosphere) • http4s (OSS)
14. PERFORMANCE
15. 3 FINCH
16. FINCH Finch is a thin layer of purely functional basic blocks on top of Finagle for building HTTP APIs. It provides developers with simple and robust HTTP primitives, while being as close as possible to the bare metal Finagle API.
17. HELLO, WORLD val service = new Service[Request, Response] { def apply(req: Request) { request.path match { case "/hello" => val resp: Response = Response() resp.content = Buf.Utf8("Hello, World!") Future.value(resp) case _ => Future.value(Response()) } } Http.server.serve(":8080", service)
18. HELLO, WORLD import io.finch._ import com.twitter.finagle.Http val api: Endpoint[String] = get("hello") { Ok("Hello, World!") } Http.server.serve(“:8080", api.toService)
19. HELLO, WORLD import io.finch._ import com.twitter.finagle.Http val api: Endpoint[String] = get("hello") { Ok("Hello, World!") } Http.server.serve(“:8080", api.toService) Finch Finagle
20. FINCH FEATURES • High level abstraction on top of Finagle (don’t need to drop down to Finagle*) • Small footprint • Flexible use (what you make of it) • Referentially transparent & compositional • Request / response decoding / encoding • Explicit async modelling
21. FINAGLE A fault tolerant, protocol-agnostic, extensible RPC system for the JVM, used to construct high- concurrency servers. Finagle implements uniform client and server APIs for several protocols, and is designed for high performance and concurrency.
22. FINAGLE FEATURES • Connection pools (w/ throttling) • Failure detection • Failover strategies • Load-balancers • Back-pressure • Statistics, logs, and exception reports • Distributed tracing (Zipkin) • Service discovery (ZooKeeper) • Sharding strategies • Config
23. TWITTERSERVER • Lightweight server template • Command line args • HTTP admin server • Logging • Tracing • Metrics • System stats
24. WHAT DOES THAT MEAN FOR YOU? • Performance & scalability out of the box • Maturity of a battle tested framework • Fast ramp up • Won’t bottom out as you scale • Known deployment, monitoring, runtime, etc.
25. 4 CORE FINCH CONCEPTS
26. TRIUMVIRATE • Endpoint • Filters • Futures • (Services)
27. ENDPOINT • A function that takes a request & returns a value • Automatically handles Future/async • Provides routing behaviour • Extracts/matches values from the request • Values are serialised to the HTTP response • Composable (applicative)
28. EXAMPLE val divOrFail: Endpoint[Int] = post("div" :: int :: int) { (a: Int, b: Int) => if (b == 0) BadRequest(new ArithmeticException("...")) else Ok(a / b) }
29. FILTER (FINAGLE) • Many common behaviours are service agnostic • Cross cutting concerns • Timeouts, logging, retries, stats, authentication, etc. • Filters are composed over services • Alter the behaviour of a service without caring what it is
30. FILTER EXAMPLE val timeout: Filter[...] val auth: Filter[...] val service: Service[Req, Resp] val composed = timeout andThen auth andThen service
31. FILTERS ARE FUNCTIONS type Filter[...] = (ReqIn, Service[ReqOut, RespIn]) => Future[RespOut]
32. FILTERS ARE TYPESAFE // A service that requires an authenticated request val service: Service[AuthReq, Resp] // Bridge with a filter val auth: Filter[HttpReq, HttpResp, AuthHttpReq, HttpResp] // Authenticate, and serve val authService: Service[HttpReq, HttpResp] = auth andThen service
33. FUTURE • A placeholder for a value that may not yet exist • Long computations, network calls, disk reads, etc. • The value is supplied concurrently (executed on thread pool) • Like callbacks, but not shit • Oh, and composable (monadic)
34. CALLBACK FUTURES val f: Future[String] f onSuccess { s => log.info(s) } onFailure { ex => log.error(ex) }
35. STATES OF A FUTURE • 3 states; empty, complete or failed • “Taints” the types of calling code • Easy to program against & make async explicit • Forces handling of async behaviour • Can also be blocked (if required)
36. FUTURE IN PRACTICE val dbUser = facebook.authenticate(token).flatMap { fbUser => val query = findByEmail(fbUser.email).result database.run(query).flatMap(_.headOption) } dbUser.transform { case Return(user) => success(user) case Throw(e) => handleError(e) }
37. SERVICE (FINAGLE) • System boundaries are represented by asynchronous functions called services • Symmetric and uniform API represents both clients and servers • You never (usually) write a Finagle service, Finch does that for you • Services are monadic (you’ll see this a lot…)
38. SERVICES ARE FUNCTIONS type Service[Req, Resp] = Req => Future[Resp]
39. SERVICES IN FINCH object LiegeApi extends ErrorOps with ResponseEncoders { private def api = usersApi() :+: ridesApi() def apiService: Service[Request, Response] = { val service = api.handle(errorHandler).toService RequestLoggingFilter.andThen(service) } }
40. 5 GOOD BITS
41. DATABINDING Given a model val ts: Endpoint[Token] = (param("t") :: param("a")).as[Token] val ts: Endpoint[Token] = Endpoint.derive[Token].fromParams case class Token(token: String, algorithm: String) Parse the querystring val getToken: Endpoint[Token] = get("tokens" ? ts) { (t: Token) => ... } Automatically parse the querystring in an endpoint
42. DATABINDING Given a model case class Token(token: String, algorithm: String) { "token": "CAAX...kfR", "algorithm": "sha1"} post("sign-in" ? body.as[Token]) { (t: Token) => ... } And incoming JSON from a POST request We can bind as
43. “CONTROLLER” object RidesApi extends HttpOps with Logging { def ridesApi() = list :+: details def list: Endpoint[List[Attendance]] = get("rides" ? authorise) { u: AuthenticatedUser => ... } def details: Endpoint[Attendance] = get("rides" / string("type") / string("id") ? authorise) { (backend: String, rideId: Id, u: AuthenticatedUser) => ... } }
44. class ItemsApiController extends Controller { val itemsService = ... val itemReader = body.as[Item] def findItemById(itemId: Long): Action = securedAction { reqContext => itemsService.findItemById(itemId) } def userItems: Action = securedAction(pageReader) { page => implicit reqContext => itemsService.userItems(user.id.get, PageRequest(page)) } override def routes: Endpoint[HttpRequest, HttpResponse] = { (Get / "api" / "items" /> userItems) | (Get / "api" / "items" / long /> findItemById) | (Post / "api" / "items" /> newItem) | } }
45. import io.finch._ import ru.arkoit.finchrich.controller._ object MyAwesomeController extends Controller { val healthcheck = get("healthcheck") { Ok() } val greeter = get("greet" / param("name")) { n: String => Ok(s"Hello, $n!") } } val ep = controllerToEndpoint(MyAwesomeController) Source: https://github.com/akozhemiakin/finchrich
46. METRICS val stats = Stats(statsReceiver) val server = Http.server.configured(stats).serve(":8081", api) val rides: Counter = statsReceiver.counter("rides") rides.incr() val ridesLatency: Stat = statsReceiver.stat("rides_latency") Stat.time(ridesLatency) { rides(u).map(rs => Ok(rs.map(r => Attendance(u, r)))) }
47. HTTP CLIENTS val client = Http.client.newService("twitter.com:8081,twitter.com:8082") val f: Future[HttpRep] = client(HttpReq("/")) val result: Future[String] = f.map { resp => handleResponse(resp) }
48. TESTING service(HttpReq("/")) map { resp => doStuff(resp) }
49. “ENTITIES” case class User( id: Option[Int] = None, name: String, email: String, location: Option[String], avatarUrl: String)
50. DDL final class UserOps(tag: Tag) extends Table[User](tag, "users") { def id = column[Int]("id", O.PrimaryKey, O.AutoInc) def name = column[String]("name") def email = column[String]("email") def location = column[String]("location") ... def nameIdx = index("name_idx", name, unique = true) }
51. QUERY OPS object UserOps extends TableQuery(new UserOps(_)) { ... }
52. TYPESAFE LOOKUPS val findByEmail = this.findBy(_.email) val findByName = this.findBy(_.name)
53. DB ACCESS def insert(u: User) = UserOps += u
54. DB ACCESS def userForToken(token: UserAccessToken): Future[Option[AuthenticatedUser]] = database.run(find(token).result).map(_.headOption.flatMap(asAuthenticatedUser)) def deauthenticateUser(token: AuthToken): Future[Unit] = { val q = for {u <- UserOps if u.authToken === token.asSessionId} yield u.authToken database.run(q.update(null)).flatMap(_ => Future.Done) }
55. MIGRATIONS object Database { lazy val migrationDatabase = new MigrationDatabase { def migrate(): Unit = { val flyway = new Flyway() flyway.setDataSource(env.dbUrl, env.dbUsername, env.dbPassword) flyway.migrate() } } }
56. 7 ISSUES
57. CONSTRAINTS • Limited support for content type negotiation • No support for schema first development (Swagger) • Only supports HTTP, JSON, XML, text, etc. (obviously)
58. ISSUES • Twitter stack • Everything is async, “sensitive to blocking code”, “reactive” bandwagon • Stuck to netty3 • Documentation not exhaustive, need to rely on Gitter
59. FUTURES • Futures are hard to compose • Representation vs. execution • Try clouds the issue
60. FUTURES • respond vs transform • respond for purely side-effecting callbacks • map & flatMap for dealing strictly with successful computations • handle and rescue for dealing strictly with exceptional computations • No support for orElse
61. 8 GETTING STARTED
62. WHEN SHOULD I USE IT? • Complex long / lived system / many developers • Scale or performance requirements • Integration with downstream services • Need to run on the JVM
63. FINCH • Watch the Finch videos • https://skillsmatter.com/skillscasts/6876-finch-your-rest- api-as-a-monad • Examples • https://github.com/finagle/finch/tree/master/examples/ src/main/scala/io/finch • Best practices • https://github.com/finagle/finch/blob/master/docs/best- practices.md
64. READ UP ON FINAGLE • Finagle Users Guide • Your function as a server (original Finagle paper) • The anatomy of a twitter microservice • Fault tolerant clients with Finagle
66. 66 QUESTIONS?
67. 7 YOW WEST SLIDES
68. IDEAS • More on request reader stuff, auth, etc. • Using circle’s auto-derivation • Abstracting/mapping Twitter Futures from/between Scala Futures from Scalaz Tasks
69. IDEAS • Other data layers • Quill, finagle-mysql