distage dist✪ge Modern Staged DependencyInjection for Scala Modular Functional Programming with Context Minimization through Garbage Collection Septimal Mind Ltd team@7mind.io 1 / 44
distage The problem: DependencyInjection and Functional Programming The motivation behind DI pattern and DI frameworks 1. Systems we work with may be represented as graphs. Nodes are components (usually instances), edges are references, 2. Graph transformation complexity grows with node count (need to add one constructor parameter, have to modify k classes), many operations have non-linear complexity, 3. Graph composition has combinatoric complexity (need to run tests choosing between mock/production repositories and external APIs, have to write four configurations). We have several possible options to address these problems: 1. Singletons and Factories: partially solve (2), code still tight coupled ⇒ expensive tests and refactorings, 2. Service Locator: bit less coupled code, still very expensive, 3. Dependency Injection: less invasive, simplifies isolation but requires more complex machinery. 2 / 44
distage The problem: DependencyInjection and Functional Programming “DI doesn’t compose with FP”: Problems 1. Typical DI framework is OOP oriented and does not support advanced concepts required for modern FP (typeclasses, higher-kinded types), 2. Almost all the DI frameworks are working in runtime while many modern FP concepts are compile-time by their nature, 3. Less guarantees: program which compiles correctly can break on wiring in runtime. After a huge delay, 4. Wiring is non-determenistic: Guice can spend several minutes trying to re-instantiate heavy instance multiple times (once per dependency) then fail, 5. Wiring is opaque: it’s hard or impossible to introspect the context. E.g. in Guice it’s a real pain to close all the instantiated Closeables. Adding missing values into the context (config injections) is not trivial as well. 3 / 44
distage The problem: DependencyInjection and Functional Programming “DI doesn’t compose with FP”: Notes 1. We have some compile-time DI frameworks or mechanisms (see MacWire) allowing us to implement DI as pattern, 2. purely compile-time tools are not convenient: sometimes we have to deal with purely runtime entities (like plugins and config values), 3. Graph composition problem is not addressed by any existing tool. 4 / 44
distage distage: Staged DIfor Scala distage: how it works DI implementations are broken. . . . . . so we may build better one, which must: 1. be well-integrated with type system of our target language (higher-kinded types, implicits, typeclasses), 2. allow us to introspect and modify our context on the fly, 3. be able to detect as many as possible problems quickly, better during compilation, 4. still allow us to deal conveniently with runtime entities, 5. give us a way to stop making atomic or conditional contexts. We made it, it works. dist✪ge 5 / 44
distage distage: Staged DIfor Scala distage: how it works Staged approach 1. Let’s apply Late Binding, 2. let’s collect our context information (bindings) first, 3. then build a DAG representing our context (so-called Project Network, let’s call it Plan), 4. then analyse this graph for errors (missing references, conflicts), 5. then apply additional transformations, 6. then interpret the graph. This is a cornercase of more generic pattern – PPER (Percept, Plan, Execute, Repeat). 6 / 44
distage distage: Staged DIfor Scala distage: how it works Staged approach: outcome What we get: 1. Planner is pure: it has no side-effects, 2. A plan is a Turing-incomplete program for a simple machine. It will always terminate in known finite time, 3. An interpreter may perform instantiations at runtime or. . . just generate Scala code that will do that when compiled, 4. All the job except for instantiations can be done in compile-time, 5. Interpreter is free to run independent instantiations in parallel, 6. Extremely important: we can transform (rewrite) the plan before we run iterpreter. 7 / 44
distage distage: Staged DIfor Scala distage: how it works Compile-Time and Runtime DI A Plan: 1 myRepository := create[MyRepository]() 2 myservice := create[MyService](myRepository) May be interpreted as: Code tree (compile-time): 1 val myRepository = 2 new MyRepository() 3 val myservice = 4 new MyService(myRepository) Set of instances (runtime): 1 plan.foldLeft(Context.empty) { 2 case (ctx, op) => 3 ctx.withInstance( 4 op.key 5 , interpret(action) 6 ) 7 } 8 / 44
distage distage: Staged DIfor Scala distage: how it works Incomplete plans This code: 1 class UsersRepoImpl(cassandraCluster: Cluster) 2 extends UsersRepo 3 class UsersService(repository: UsersRepo) 4 5 class UsersModule extends ModuleDef { 6 make[UsersRepo].from[UsersRepoImpl] 7 make[UsersService] 8 } May produce a plan like: 1 cassandraCluster := import[Cluster] 2 usersRepo: UsersRepo := create[UsersRepoImpl](cassandraCluster) 3 usersService := create[UsersService](usersRepo) 9 / 44
distage distage: Staged DIfor Scala patterns and Extensions Pattern: Plan completion Once we have such a plan: 1 cassandraCluster := import[Cluster] 2 usersRepo: UsersRepo := create[UsersRepoImpl](cassandraCluster) 3 usersService := create[UsersService](usersRepo) We may add missing values1: 1 val plan = Injector.plan(definitions) 2 val resolved = plan.map { 3 case i: Import if i.is[Cluster] => 4 val cluster: Cluster = ??? 5 Reference(cluster) 6 case op => op 7 } 1 Pseudocode, real API is bit different 11 / 44
distage distage: Staged DIfor Scala patterns and Extensions Extension: Configuration Support distage has HOCON configuration support implemented as an extension. 1 case class HostPort(host: String, port: Int) 2 3 class HttpServer(@ConfPath("http.listen") listenOn: HostPort) { 4 // ... 5 } The extension: 1. Takes all the Imports of a Plan, 2. Searches them for a specific @ConfPath annotation, 3. Tries to find corresponding sections in config, 4. Extends plan with config values, All the config values are resolved even before instantiation of services ⇒ problems are being shown quickly and all at once. 12 / 44
distage distage: Staged DIfor Scala patterns and Extensions Extension: Automatic Sets 1. distage can find all instances type T (like AutoCloseable) in the context, put them all into a Set[T] then inject that set. 2. ⇒ basic lifecycle support, free of charge. 1 trait Resource { 2 def start(): Unit 3 def stop(): Unit 4 } 5 trait App { def main(): Unit } 6 locator.run { (resources: Set[Resource], app: App) => 7 try { 8 resources.foreach(_.start()) 9 app.main() 10 } finally { 11 resources.foreach(_.close()) 12 } 13 } 13 / 44
distage distage: Staged DIfor Scala patterns and Extensions The Principle Behind: Feedback Loops They use so-called Feedback Loops in robotics. . . 14 / 44
distage distage: Staged DIfor Scala patterns and Extensions The Principle Behind: PPER Loop1 We found a very generic and important pattern of Feedback Loop class: 1. Acquire data from the outer world (Percept) 2. Produce a Project Network, Plan. It may be incomplete, but should allow us to progress (Plan) ◮ Plan is a DAG, actions are nodes, edges are dependencies 3. Execute the Plan (Execute). ◮ Perform the steps of the Plan ◮ Mark your Plan nodes according to the results of their execution ◮ Let’s call marked plan as Trace 4. Go to step 1 unless termination criteria reached (Repeat) Let’s call it PPER (pronounced: pepper). 1 Slides: https://goo.gl/okZ8Bw 15 / 44
distage distage: Staged DIfor Scala Garbage Collector and its Benefits Garbage Collector and Context Minimization 1. Let’s assume that we have a UsersService and AccountingService in your context, 2. . . . and we want to write a test for UsersService only, 3. We may exploit staged design and collect the garbage out of Plan before executing it. 4. We define a garbage collection root, UsersService, and keep only the operations it transitively depends on. The rest is being thrown out even before it’s being instantiated, 5. Garbage Collector allows us to compose contexts easier. 16 / 44
distage distage: Staged DIfor Scala Garbage Collector and its Benefits Garbage Collector and Context Minimization 17 / 44
distage distage: Staged DIfor Scala Garbage Collector and its Benefits Context Minimization for Tests Context minimization allows us to: 1. Instantiate only the instances which are required for your tests, 2. Save on test startup time (savings may be significant), 3. Save on configuring per-test contexts manually (savings may be substantial). 18 / 44
distage distage: Staged DIfor Scala Garbage Collector and its Benefits Context Minimization for Deployment Context minimization allows us to: 1. Have one image with all our software components (Roles1), 2. . . . keeping development flows of these components isolated, 3. Decide which components we want to run when we start the image, 4. Have higher computational density 5. substantially simplify development flows: we may run full environment with a single command on a low-end machine, 6. Fuse Microservices with Monoliths keeping all their benefits. 1 server1# docker run -ti company/product +analytics 2 server2# docker run -ti company/product +accounting +users 3 laptop# docker run -ti company/product --run-everything 1 Slides: https://goo.gl/iaMt43 19 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming dist✪ge Scala Typesystem Integration: Fusional Programming 20 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming Kind-Polymorphic Type Tags ◮ distage uses types as lookup keys ◮ But scalac has TypeTags only for kind "*"! ◮ distage added any-kinded ‘Tag‘s ◮ Modules can be polymorphic on types of any kind 1 def m1[F[_]: TagK: Monad, A: Encoder] 2 = new ModuleDef { 3 make[SendJson[F, A]] 4 addImplicit[Encoder[A]] 5 addImplicit[Monad[F]] 6 make[Network[F]].from(...) } 7 // `Tag` implicit accepts 8 // type parameters of arbitrary kind 9 def mTrans[FT[_[_], _]: Tag.auto.T] = 10 m1[FT[IO], AccountReport] 11 mTrans[OptionT] 1 type TagK[F[_]] = HKTag[{ type Arg[A] = F[A] }] 2 type TagKK[F[_, _]] = HKTag[{ type Arg[A, B] = F[A, B] }] 3 type Tag.auto.T[F[...]] = HKTag[{ type Arg[...] = F[...] }] 21 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming Typeclass instance injection (Implicit Injection) 1. distage wires all parameters, inlcuding implicits 2. Leaving implicits solely to implicit resolution would’ve been unmodular, because we want to bind instances late, not at definition time 3. Therefore implicits should be declared explicitly 4. Tagless-final style is encouraged – add implicit parameters, instead of importing concrete implicits! 1 def polyModule[F[_]: Sync: TagK] = new ModuleDef { 2 make[Sync[F]].from(implicitly[Sync[F]]) 3 make[F[Unit]].named("hello").from( 4 S: Sync[F] => S.delay(println("Hello World!"))) 5 } 6 Injector() 7 .produce(polyModule[IO]) 8 .get[IO[Unit]]("hello").unsafeRunSync() 22 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming Code example: Tagless Final Style 1 trait Interaction[F[_]] { 2 def say(msg: String): F[Unit] 3 def ask(prompt: String): F[String] } 4 5 class TaglessHelloWorld[F[_]: Monad: Interaction] { 6 val program = for { 7 userName <- Interaction[F].ask("What's your name?") 8 _ <- Interaction[F].say(s"Hello $userName!") 9 } yield () } 10 11 def wiring[F[_]: TagK: Sync] = new ModuleDef { 12 make[TaglessHelloWorld[F]] 13 make[Interaction[F]].from(new Interaction[F] {...}) 14 addImplicit[Monad[F]] } 15 16 Injector().produce(wiring[IO]).get[TaglessHelloWorld[IO]] 17 .program.unsafeRunSync() 23 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming Code example: Scalaz ZIO Injection 1 trait MonadFilesystem[F[_, _]] { 2 def readFile(fileName: String): F[Exception, String] 3 } 4 5 class ZioFilesystemModule extends ModuleDef { 6 make[MonadFilesystem[IO]].from(new MonadFilesystem[IO] { 7 def readFile(fileName: String) = 8 IO.syncException(Source.fromFile(fileName)("UTF-8") 9 .mkString) 10 }) 11 make[GetCpus[IO]] 12 } 13 14 class GetCpus(fs: MonadFilesystem[IO]) { 15 def apply(): IO[Exception, Int] = 16 fs.readFile("/proc/cpuinfo").flatMap(...) 17 } 24 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming Lambda Injection, ProviderMagnet 1. We can bind functions as constructors 2. Arbitrary-arity functions can run with arguments from Locator 3. Put annotations on types to summon named instances or config values (from distage-config) 1 case class A() 2 case class B(a: A, default: Int = 5) 3 val ctx: Locator = Injector(config).produce(new ModuleDef { 4 make[A].from(A()) 5 make[B].from{ a: A => B(a) } 6 }) 7 case class C(a: A, b: B, host: HostPort) 8 val c: C = ctx.run { 9 (a: A, b: B, host: HostPort @ConfPath("http.listen")) => 10 C(a, b, host) 11 } 25 / 44
distage distage: Staged DIfor Scala Scala Typesystem Integration: Fusional Programming Path-dependent types 1. Path-dependent types and projections are supported 2. Path-prefix should be wired for the child type to be wired 1 trait Cake { trait Child } 2 3 val cakeModule = new ModuleDef { 4 make[Cake] 5 make[Cake#Child] 6 } 7 val cake = Injector().produce(cakeModule).get[Cake] 8 9 val instanceCakeModule = new ModuleDef { 10 make[cake.type].from[cake.type](cake: cake.type) 11 make[cake.Child] 12 } 13 val cakeChild = Injector().produce(instanceCakeModule) 14 .get[cake.Child] 26 / 44
distage distage: Staged DIfor Scala Features to boost productivity dist✪ge Features to boost productivity 27 / 44
distage distage: Staged DIfor Scala Features to boost productivity Dynamic Plugins Just drop your modules into your classpath: 1 class AccountingModule extends PluginDef { 2 make[AccountingService].from[AccountingServiceImpl] 3 // ... 4 } Then you may pick up all the modules and build your context: 1 val plugins = new PluginLoaderDefaultImpl( 2 PluginConfig(Seq("com.company.plugins")) 3 ).load() 4 // ... pass to an Injector 1. Useful while you are prototyping your app, 2. In maintenance phase you may switch to static configuration. 28 / 44
distage distage: Staged DIfor Scala Features to boost productivity Dynamic Testkit (ScalaTest) 1 class AccountingServiceTest extends DistagePluginSpec { 2 "accounting service" must { 3 "be resolved dynamically" in di { 4 (acc: AccountingService) => 5 assert(acc.getBalance("bad-user") == 0) 6 } 7 } 8 } 1. You don’t need to setup your context, it’s done automatically by Plugin Loader and Garbage Collector, 2. And it takes just milliseconds, not like in Spring, 3. Garbage collection roots are inferred from test’s signature, 4. Only the things required for a particular test are being instantiated. 29 / 44
distage distage: Staged DIfor Scala Features to boost productivity Tags 1. Each binding may be marked with a tag, 2. Some bindings may be excluded from the context before planning by a predicate, 3. This is unsafe but convenient way to reconfigure contexts conditionally. 1 class ProductionPlugin extends PluginDef { 2 tag("production", "cassandra") 3 make[UserRepo].from[CassandraUserRepo] 4 } 5 class MockPlugin extends PluginDef { 6 tag("test", "mock") 7 make[UserRepo].from[MockUserRepo] 8 } 9 // ... 10 val disabledTags = t"mock" && t"dummy" 11 val plan = injector.plan(definition.filter(disabledTags)) 30 / 44
distage distage: Staged DIfor Scala Features to boost productivity Circular dependencies 1. Supported, Proxy concept used, 2. By-name parameters (class C(param: => P)) supported, no runtime code-generation requried, 3. Other cases are supported as well with runtime code-generation, 4. Limitations: typical. You cannot use an injected parameter immediately in a constructor, you cannot have circular non-by-name dependencies with final classes, 5. Circular dependency resolution is optional, you may turn it off. 31 / 44
distage distage: Staged DIfor Scala Features to boost productivity Plan Introspection: example context 1 class Cluster 2 trait UsersService 3 trait AccountingService 4 trait UserRepo 5 trait AccountsRepo 6 7 class UserRepoImpl(cluster: Cluster) extends UserRepo 8 class AccountsRepoImpl(cluster: Cluster) extends AccountsRepo 9 class UserServiceImpl(userRepo: UserRepo) extends UsersService 10 class AccountingServiceImpl(accountsRepo: AccountsRepo) 11 extends AccountingService 12 13 class UsersApiImpl(service: UsersService 14 , accountsApi: AccountsApiImpl) 15 class AccountsApiImpl(service: AccountingService 16 , usersApi: UsersApiImpl) // circular dependency 17 class App(uapi: UsersApiImpl, aapi: AccountsApiImpl) 32 / 44
distage distage: Staged DIfor Scala Features to boost productivity Plan Introspection: example bindings1 1 val definition = new ModuleDef { 2 make[Cluster] 3 make[UserRepo].from[UserRepoImpl] 4 make[AccountsRepo].from[AccountsRepoImpl] 5 make[UsersService].from[UserServiceImpl] 6 make[AccountingService].from[AccountingServiceImpl] 7 make[UsersApiImpl] 8 make[AccountsApiImpl] 9 make[App] 10 } 11 val injector = Injector() 12 val plan = injector.plan(definition) 1 Full code example: https://goo.gl/7ZwHfX 33 / 44
distage distage: Staged DIfor Scala Features to boost productivity Plan Introspection: plan dumps 1 println(plan.render) // look for the circular dependency! 34 / 44
distage distage: Staged DIfor Scala Features to boost productivity Plan Introspection: dependency trees You may explore dependencies of a component: 1 val dependencies = plan.topology.dependencies 2 println(dependencies.tree(DIKey.get[AccountsApiImpl])) Circular dependencies are specifically marked. 35 / 44
distage distage: Staged DIfor Scala Features to boost productivity Trait Completion 1 trait UsersService { 2 protected def repository: UsersRepo 3 def add(user: User): Unit = { 4 repository.put(user.id, user) 5 ??? 6 } 7 } We may bind this trait directly, without an implementation class: 1 make[UsersService] 1. Corresponding class will be generated by distage, 2. Null-arg abstract methods will be wired with context values, 3. Works in both runtime and compile-time. 36 / 44
distage distage: Staged DIfor Scala Features to boost productivity Factory Methods (Assisted Injection) 1 class UserActor(sessionId: UUID, sessionRepo: SessionRepo) 2 3 trait ActorFactory { 4 def createActor(sessionId: UUID): UserActor 5 } 1. createActor is a factory method, 2. createActor will be generated by distage, 3. non-null-arg abstract methods are treated as factory methods, 4. Non-invasive assisted injection: sessionId: UUID will be taken from method parameter, sessionRepo: SessionRepo will be wired from context, 5. Useful for Akka, lot more convenient than Guice, 6. Works in both runtime and compile-time. 37 / 44
distage distage: Staged DIfor Scala 7mind Stack distage: status and things to do distage is: 1. ready to use, 2. in real production, 3. all Runtime APIs are available, 4. Compile-time verification, trait completion, assisted injections and lambda injections are available. Our plans: 1. Refactor Roles API, 2. Support running Producer within a monad (to use with Scalaz ZIO, Monix, cats-effect, etc), 3. Support Scala.js, 4. Support optional isolated classloaders (in foreseeable future), 5. Publish compile-time Producer, 6. Check our GitHub: https://github.com/pshirshov/izumi-r2. 39 / 44
distage distage: Staged DIfor Scala 7mind Stack distage is just a part of our stack We have a vision backed by our tools: 1. Idealingua: transport and codec agnostic gRPC alternative with rich modeling language, 2. LogStage: zero-cost logging framework, 3. Fusional Programming and Design guidelines. We love both FP and OOP, 4. Continous Delivery guidelines for Role-based process, 5. Percept-Plan-Execute Generative Programming approach, abstract machine and computational model. Addresses Project Planning (see Operations Research). Examples: orchestration, build systems. Altogether these things already allowed us to significantly reduce development costs and delivery time for our client. More slides to follow. 40 / 44
distage distage: Staged DIfor Scala 7mind Stack Teaser: LogStage A log call . . . 1 log.info(s"$user logged in with $sessionId!") . . . may be rendered as a text like 17:05:18 UserService.login user=John Doe logged in with sessionId=DEADBEEF! . . . or a structured JSON: 1 { 2 "user": "John Doe", 3 "sessionId": "DEADBEEF", 4 "_template": "$user logged in with $sessionId!", 5 "_location": "UserService.scala:265", 6 "_context": "UserService.login", 7 } 42 / 44
distage distage: Staged DIfor Scala 7mind Stack Teaser: Idealingua 1 id UserId { uid: str } 2 data User { name: str /* ... more fields */ } 3 data PublicUser { 4 + InternalUser 5 - SecurityAttributes 6 } 7 adt Failure = NotFound | UnknownFailure 8 service Service { 9 def getUser(id: UserId): User !! Failure 10 } 1. Convenient Data and Interface Definition Language, 2. Extensible, transport-agnostic, abstracted from wire format, 3. JSON + HTTP / WebSocket at the moment, 4. C#, go, Scala, TypeScript at the moment, 5. Better than gRPC/ REST / Swagger/ etc. 43 / 44
distage distage: Staged DIfor Scala 7mind Stack Thank you for your attention distage website: https://izumi.7mind.io/ We’re looking for clients, contributors, adopters and colleagues ;) About the author: 1. coding for 18 years, 10 years of hands-on commercial engineering experience, 2. has been leading a cluster orchestration team in Yandex, “the Russian Google”, 3. implemented “Interstellar Spaceship” – an orchestration solution to manage 50K+ physical machines across 6 datacenters, 4. Owns an Irish R&D company, https://7mind.io, 5. Contact: team@7mind.io, 6. Github: https://github.com/pshirshov 7. Download slides: https://github.com/7mind/slides/ 44 / 44