Sistemi reattivi in ​​Java

1. Introduzione

In questo tutorial, comprenderemo le basi della creazione di sistemi reattivi in ​​Java utilizzando Spring e altri strumenti e framework.

Nel processo, discuteremo di come la programmazione reattiva sia solo un driver verso la creazione di un sistema reattivo. Questo ci aiuterà a comprendere la logica alla base della creazione di sistemi reattivi e diverse specifiche, librerie e standard che ha ispirato lungo il percorso.

2. Cosa sono i sistemi reattivi?

Negli ultimi decenni, il panorama tecnologico ha subito diverse interruzioni che hanno portato a una trasformazione completa nel modo in cui vediamo il valore nella tecnologia. Il mondo dei computer prima di Internet non avrebbe mai potuto immaginare i modi e i mezzi in cui cambierà i nostri giorni.

Con la portata di Internet alle masse e l'esperienza in continua evoluzione che promette, gli architetti delle applicazioni devono essere all'erta per soddisfare la loro domanda.

Fondamentalmente, questo significa che non possiamo mai progettare un'applicazione nel modo in cui facevamo prima. Un'applicazione altamente reattiva non è più un lusso ma una necessità .

Anche questo è di fronte a guasti casuali e carichi imprevedibili. Il bisogno dell'ora non è solo quello di ottenere il risultato corretto, ma di ottenerlo velocemente! È molto importante promuovere le straordinarie esperienze utente che promettiamo di offrire.

Questo è ciò che crea la necessità di uno stile architettonico che possa darci sistemi reattivi.

2.1. Manifesto reattivo

Nel 2013, un team di sviluppatori, guidato da Jonas Boner, si è riunito per definire una serie di principi fondamentali in un documento noto come Reactive Manifesto. Questo è ciò che ha gettato le basi per uno stile di architettura per creare sistemi reattivi. Da allora, questo manifesto ha raccolto molto interesse da parte della comunità degli sviluppatori.

Fondamentalmente, questo documento prescrive la ricetta per un sistema reattivo che sia flessibile, liberamente accoppiato e scalabile . Ciò rende tali sistemi facili da sviluppare, tolleranti ai guasti e, cosa più importante, altamente reattivi, alla base di incredibili esperienze utente.

Allora qual è questa ricetta segreta? Beh, non è quasi un segreto! Il manifesto definisce le caratteristiche oi principi fondamentali di un sistema reattivo:

  • Reattivo : un sistema reattivo dovrebbe fornire un tempo di risposta rapido e coerente e quindi una qualità del servizio costante
  • Resiliente : un sistema reattivo dovrebbe rimanere reattivo in caso di guasti casuali attraverso la replica e l'isolamento
  • Elastico : un sistema di questo tipo dovrebbe rimanere reattivo in presenza di carichi di lavoro imprevedibili grazie a una scalabilità conveniente
  • Basato sui messaggi : dovrebbe fare affidamento sul passaggio di messaggi asincrono tra i componenti del sistema

Questi principi sembrano semplici e sensati, ma non sono sempre più facili da implementare in un'architettura aziendale complessa. In questo tutorial, svilupperemo un sistema di esempio in Java con questi principi in mente!

3. Che cos'è la programmazione reattiva?

Prima di procedere, è importante comprendere la differenza tra programmazione reattiva e sistemi reattivi. Usiamo entrambi questi termini abbastanza spesso e facilmente fraintendiamo l'uno per l'altro. Come abbiamo visto in precedenza, i sistemi reattivi sono il risultato di uno stile architettonico specifico.

Al contrario, la programmazione reattiva è un paradigma di programmazione in cui l'attenzione è rivolta allo sviluppo di componenti asincroni e non bloccanti . Il nucleo della programmazione reattiva è un flusso di dati a cui possiamo osservare e reagire, anche applicare la contropressione. Ciò porta a un'esecuzione non bloccante e quindi a una migliore scalabilità con meno thread di esecuzione.

Ora, questo non significa che i sistemi reattivi e la programmazione reattiva si escludano a vicenda. In effetti, la programmazione reattiva è un passo importante verso la realizzazione di un sistema reattivo, ma non è tutto!

3.1. Stream reattivi

Reactive Streams è un'iniziativa della comunità iniziata nel 2013 per fornire uno standard per l'elaborazione di flussi asincroni con contropressione non bloccante . L'obiettivo qui era definire una serie di interfacce, metodi e protocolli in grado di descrivere le operazioni e le entità necessarie.

Da allora, sono emerse diverse implementazioni in più linguaggi di programmazione conformi alla specifica dei flussi reattivi. Questi includono Akka Streams, Ratpack e Vert.x per citarne alcuni.

3.2. Librerie reattive per Java

Uno degli obiettivi iniziali dietro i flussi reattivi era quello di essere eventualmente incluso come libreria standard Java ufficiale. Di conseguenza, la specifica dei flussi reattivi è semanticamente equivalente alla libreria Java Flow, introdotta in Java 9.

A parte questo, ci sono alcune scelte popolari per implementare la programmazione reattiva in Java:

  • Estensioni reattive: popolarmente note come ReactiveX, forniscono API per la programmazione asincrona con flussi osservabili. Questi sono disponibili per più linguaggi di programmazione e piattaforme, incluso Java dove è noto come RxJava
  • Project Reactor: questa è un'altra libreria reattiva, basata sulla specifica dei flussi reattivi, mirata alla creazione di non applicazioni sulla JVM. Capita anche che sia il fondamento dello stack reattivo nell'ecosistema primaverile

4. Una semplice applicazione

Ai fini di questo tutorial, svilupperemo una semplice applicazione basata su un'architettura di microservizi con un frontend minimo. L'architettura dell'applicazione dovrebbe avere elementi sufficienti per creare un sistema reattivo.

Per la nostra applicazione, adotteremo una programmazione reattiva end-to-end e altri modelli e strumenti per realizzare le caratteristiche fondamentali di un sistema reattivo.

4.1. Architettura

Inizieremo definendo una semplice architettura applicativa che non presenta necessariamente le caratteristiche dei sistemi reattivi . Da lì in poi, apporteremo le modifiche necessarie per ottenere queste caratteristiche una per una.

Quindi, per prima cosa, iniziamo definendo una semplice architettura:

Questa è un'architettura abbastanza semplice che ha un sacco di microservizi per facilitare un caso d'uso commerciale in cui possiamo effettuare un ordine. Ha anche un frontend per l'esperienza utente e tutte le comunicazioni avvengono come REST su HTTP. Inoltre, ogni microservizio gestisce i propri dati in singoli database, una pratica nota come database per servizio.

Andremo avanti e creeremo questa semplice applicazione nelle seguenti sottosezioni. Questa sarà la nostra base per comprendere gli errori di questa architettura e modi e mezzi per adottare principi e pratiche in modo da poterlo trasformare in un sistema reattivo.

4.3. Microservizio di inventario

Inventory microservice will be responsible for managing a list of products and their current stock. It will also allow altering the stock as orders are processed. We'll use Spring Boot with MongoDB to develop this service.

Let's begin by defining a controller to expose some endpoints:

@GetMapping public List getAllProducts() { return productService.getProducts(); } @PostMapping public Order processOrder(@RequestBody Order order) { return productService.handleOrder(order); } @DeleteMapping public Order revertOrder(@RequestBody Order order) { return productService.revertOrder(order); }

and a service to encapsulate our business logic:

@Transactional public Order handleOrder(Order order) { order.getLineItems() .forEach(l -> { Product> p = productRepository.findById(l.getProductId()) .orElseThrow(() -> new RuntimeException("Could not find the product: " + l.getProductId())); if (p.getStock() >= l.getQuantity()) { p.setStock(p.getStock() - l.getQuantity()); productRepository.save(p); } else { throw new RuntimeException("Product is out of stock: " + l.getProductId()); } }); return order.setOrderStatus(OrderStatus.SUCCESS); } @Transactional public Order revertOrder(Order order) { order.getLineItems() .forEach(l -> { Product p = productRepository.findById(l.getProductId()) .orElseThrow(() -> new RuntimeException("Could not find the product: " + l.getProductId())); p.setStock(p.getStock() + l.getQuantity()); productRepository.save(p); }); return order.setOrderStatus(OrderStatus.SUCCESS); }

Note that we're persisting the entities within a transaction, which ensures that no inconsistent state results in case of exceptions.

Apart from these, we'll also have to define the domain entities, the repository interface, and a bunch of configuration classes necessary for everything to work properly.

But since these are mostly boilerplate, we'll avoid going through them, and they can be referred to in the GitHub repository provided in the last section of this article.

4.4. Shipping Microservice

The shipping microservice will not be very different either. This will be responsible for checking if a shipment can be generated for the order and create one if possible.

As before we'll define a controller to expose our endpoints, in fact just a single endpoint:

@PostMapping public Order process(@RequestBody Order order) { return shippingService.handleOrder(order); }

and a service to encapsulate the business logic related to order shipment:

public Order handleOrder(Order order) { LocalDate shippingDate = null; if (LocalTime.now().isAfter(LocalTime.parse("10:00")) && LocalTime.now().isBefore(LocalTime.parse("18:00"))) { shippingDate = LocalDate.now().plusDays(1); } else { throw new RuntimeException("The current time is off the limits to place order."); } shipmentRepository.save(new Shipment() .setAddress(order.getShippingAddress()) .setShippingDate(shippingDate)); return order.setShippingDate(shippingDate) .setOrderStatus(OrderStatus.SUCCESS); }

Our simple shipping service is just checking the valid time window to place orders. We'll avoid discussing the rest of the boilerplate code as before.

4.5. Order Microservice

Finally, we'll define an order microservice which will be responsible for creating a new order apart from other things. Interestingly, it'll also play as an orchestrator service where it will communicate with the inventory service and the shipping service for the order.

Let's define our controller with the required endpoints:

@PostMapping public Order create(@RequestBody Order order) { Order processedOrder = orderService.createOrder(order); if (OrderStatus.FAILURE.equals(processedOrder.getOrderStatus())) { throw new RuntimeException("Order processing failed, please try again later."); } return processedOrder; } @GetMapping public List getAll() { return orderService.getOrders(); }

And, a service to encapsulate the business logic related to orders:

public Order createOrder(Order order) { boolean success = true; Order savedOrder = orderRepository.save(order); Order inventoryResponse = null; try { inventoryResponse = restTemplate.postForObject( inventoryServiceUrl, order, Order.class); } catch (Exception ex) { success = false; } Order shippingResponse = null; try { shippingResponse = restTemplate.postForObject( shippingServiceUrl, order, Order.class); } catch (Exception ex) { success = false; HttpEntity deleteRequest = new HttpEntity(order); ResponseEntity deleteResponse = restTemplate.exchange( inventoryServiceUrl, HttpMethod.DELETE, deleteRequest, Order.class); } if (success) { savedOrder.setOrderStatus(OrderStatus.SUCCESS); savedOrder.setShippingDate(shippingResponse.getShippingDate()); } else { savedOrder.setOrderStatus(OrderStatus.FAILURE); } return orderRepository.save(savedOrder); } public List getOrders() { return orderRepository.findAll(); }

The handling of orders where we're orchestrating calls to inventory and shipping services is far from ideal. Distributed transactions with multiple microservices is a complex topic in itself and beyond the scope of this tutorial.

However, we'll see later in this tutorial how a reactive system can avoid the need for distributed transactions to a certain extent.

As before, we'll not go through the rest of the boilerplate code. However, this can be referenced in the GitHub repo.

4.6. Front-end

Let's also add a user interface to make the discussion complete. The user interface will be based on Angular and will be a simple single-page application.

We'll need to create a simple component in Angular to handle create and fetch orders. Of specific importance is the part where we call our API to create the order:

createOrder() { let headers = new HttpHeaders({'Content-Type': 'application/json'}); let options = {headers: headers} this.http.post('//localhost:8080/api/orders', this.form.value, options) .subscribe( (response) => { this.response = response }, (error) => { this.error = error } ) }

The above code snippet expects order data to be captured in a form and available within the scope of the component. Angular offers fantastic support for creating simple to complex forms using reactive and template-driven forms.

Also important is the part where we get previously created orders:

getOrders() { this.previousOrders = this.http.get(''//localhost:8080/api/orders'') }

Please note that the Angular HTTP module is asynchronous in nature and hence returns RxJS Observables. We can handle the response in our view by passing them through an async pipe:

Your orders placed so far:

  • Order ID: {{ order.id }}, Order Status: {{order.orderStatus}}, Order Message: {{order.responseMessage}}

Of course, Angular will require templates, styles, and configurations to work, but these can be referred to in the GitHub repository. Please note that we have bundled everything in a single component here, which is ideally not something we should do.

But, for this tutorial, those concerns are not in scope.

4.7. Deploying the Application

Now that we've created all individual parts of the application, how should we go about deploying them? Well, we can always do this manually. But we should be careful that it can soon become tedious.

For this tutorial, we'll use Docker Compose to build and deploy our application on a Docker Machine. This will require us to add a standard Dockerfile in each service and create a Docker Compose file for the entire application.

Let's see how this docker-compose.yml file looks:

version: '3' services: frontend: build: ./frontend ports: - "80:80" order-service: build: ./order-service ports: - "8080:8080" inventory-service: build: ./inventory-service ports: - "8081:8081" shipping-service: build: ./shipping-service ports: - "8082:8082"

This is a fairly standard definition of services in Docker Compose and does not require any special attention.

4.8. Problems With This Architecture

Now that we have a simple application in place with multiple services interacting with each other, we can discuss the problems in this architecture. There are what we'll try to address in the following sections and eventually get to the state where we would have transformed our application into a reactive system!

While this application is far from a production-grade software and there are several issues, we'll focus on the issues that pertain to the motivations for reactive systems:

  • Failure in either inventory service or shipping service can have a cascading effect
  • The calls to external systems and database are all blocking in nature
  • The deployment cannot handle failures and fluctuating loads automatically

5. Reactive Programming

Blocking calls in any program often result in critical resources just waiting for things to happen. These include database calls, calls to web services, and file system calls. If we can free up threads of execution from this waiting and provide a mechanism to circle back once results are available, it will yield much better resource utilization.

This is what adopting the reactive programming paradigm does for us. While it's possible to switch over to a reactive library for many of these calls, it may not be possible for everything. For us, fortunately, Spring makes it much easier to use reactive programming with MongoDB and REST APIs:

Spring Data Mongo has support for reactive access through the MongoDB Reactive Streams Java Driver. It provides ReactiveMongoTemplate and ReactiveMongoRepository, both of which have extensive mapping functionality.

Spring WebFlux provides the reactive-stack web framework for Spring, enabling non-blocking code and Reactive Streams backpressure. It leverages the Reactor as its reactive library. Further, it provides WebClient for performing HTTP requests with Reactive Streams backpressure. It uses Reactor Netty as the HTTP client library.

5.1. Inventory Service

We'll begin by changing our endpoints to emit reactive publishers:

@GetMapping public Flux getAllProducts() { return productService.getProducts(); }
@PostMapping public Mono processOrder(@RequestBody Order order) { return productService.handleOrder(order); } @DeleteMapping public Mono revertOrder(@RequestBody Order order) { return productService.revertOrder(order); }

Obviously, we'll have to make necessary changes to the service as well:

@Transactional public Mono handleOrder(Order order) { return Flux.fromIterable(order.getLineItems()) .flatMap(l -> productRepository.findById(l.getProductId())) .flatMap(p -> { int q = order.getLineItems().stream() .filter(l -> l.getProductId().equals(p.getId())) .findAny().get() .getQuantity(); if (p.getStock() >= q) { p.setStock(p.getStock() - q); return productRepository.save(p); } else { return Mono.error(new RuntimeException("Product is out of stock: " + p.getId())); } }) .then(Mono.just(order.setOrderStatus("SUCCESS"))); } @Transactional public Mono revertOrder(Order order) { return Flux.fromIterable(order.getLineItems()) .flatMap(l -> productRepository.findById(l.getProductId())) .flatMap(p -> { int q = order.getLineItems().stream() .filter(l -> l.getProductId().equals(p.getId())) .findAny().get() .getQuantity(); p.setStock(p.getStock() + q); return productRepository.save(p); }) .then(Mono.just(order.setOrderStatus("SUCCESS"))); }

5.2. Shipping Service

Similarly, we'll change the endpoint of our shipping service:

@PostMapping public Mono process(@RequestBody Order order) { return shippingService.handleOrder(order); }

And, corresponding changes in the service to leverage reactive programming:

public Mono handleOrder(Order order) { return Mono.just(order) .flatMap(o -> { LocalDate shippingDate = null; if (LocalTime.now().isAfter(LocalTime.parse("10:00")) && LocalTime.now().isBefore(LocalTime.parse("18:00"))) { shippingDate = LocalDate.now().plusDays(1); } else { return Mono.error(new RuntimeException("The current time is off the limits to place order.")); } return shipmentRepository.save(new Shipment() .setAddress(order.getShippingAddress()) .setShippingDate(shippingDate)); }) .map(s -> order.setShippingDate(s.getShippingDate()) .setOrderStatus(OrderStatus.SUCCESS)); }

5.3. Order Service

We'll have to make similar changes in the endpoints of the order service:

@PostMapping public Mono create(@RequestBody Order order) { return orderService.createOrder(order) .flatMap(o -> { if (OrderStatus.FAILURE.equals(o.getOrderStatus())) { return Mono.error(new RuntimeException("Order processing failed, please try again later. " + o.getResponseMessage())); } else { return Mono.just(o); } }); } @GetMapping public Flux getAll() { return orderService.getOrders(); }

The changes to service will be more involved as we'll have to make use of Spring WebClient to invoke the inventory and shipping reactive endpoints:

public Mono createOrder(Order order) { return Mono.just(order) .flatMap(orderRepository::save) .flatMap(o -> { return webClient.method(HttpMethod.POST) .uri(inventoryServiceUrl) .body(BodyInserters.fromValue(o)) .exchange(); }) .onErrorResume(err -> { return Mono.just(order.setOrderStatus(OrderStatus.FAILURE) .setResponseMessage(err.getMessage())); }) .flatMap(o -> { if (!OrderStatus.FAILURE.equals(o.getOrderStatus())) { return webClient.method(HttpMethod.POST) .uri(shippingServiceUrl) .body(BodyInserters.fromValue(o)) .exchange(); } else { return Mono.just(o); } }) .onErrorResume(err -> { return webClient.method(HttpMethod.POST) .uri(inventoryServiceUrl) .body(BodyInserters.fromValue(order)) .retrieve() .bodyToMono(Order.class) .map(o -> o.setOrderStatus(OrderStatus.FAILURE) .setResponseMessage(err.getMessage())); }) .map(o -> { if (!OrderStatus.FAILURE.equals(o.getOrderStatus())) { return order.setShippingDate(o.getShippingDate()) .setOrderStatus(OrderStatus.SUCCESS); } else { return order.setOrderStatus(OrderStatus.FAILURE) .setResponseMessage(o.getResponseMessage()); } }) .flatMap(orderRepository::save); } public Flux getOrders() { return orderRepository.findAll(); }

This kind of orchestration with reactive APIs is no easy exercise and often error-prone as well as hard to debug. We'll see how this can be simplified in the next section.

5.4. Front-end

Now, that our APIs are capable of streaming events as they occur, it's quite natural that we should be able to leverage that in our front-end as well. Fortunately, Angular supports EventSource, the interface for Server-Sent Events.

Let's see how can we pull and process all our previous orders as a stream of events:

getOrderStream() { return Observable.create((observer) => { let eventSource = new EventSource('//localhost:8080/api/orders') eventSource.onmessage = (event) => { let json = JSON.parse(event.data) this.orders.push(json) this._zone.run(() => { observer.next(this.orders) }) } eventSource.onerror = (error) => { if(eventSource.readyState === 0) { eventSource.close() this._zone.run(() => { observer.complete() }) } else { this._zone.run(() => { observer.error('EventSource error: ' + error) }) } } }) }

6. Message-Driven Architecture

The first problem we're going to address is related to service-to-service communication. Right now, these communications are synchronous, which presents several problems. These include cascading failures, complex orchestration, and distributed transactions to name a few.

An obvious way to solve this problem is to make these communications asynchronous. A message broker for facilitating all service-to-service communication can do the trick for us. We'll use Kafka as our message broker and Spring for Kafka to produce and consume messages:

We'll use a single topic to produce and consume order messages with different order statuses for services to react.

Let's see how each service needs to change.

6.1. Inventory Service

Let's begin by defining the message producer for our inventory service:

@Autowired private KafkaTemplate kafkaTemplate; public void sendMessage(Order order) { this.kafkaTemplate.send("orders", order); }

Next, we'll have to define a message consumer for inventory service to react to different messages on the topic:

@KafkaListener(topics = "orders", groupId = "inventory") public void consume(Order order) throws IOException { if (OrderStatus.RESERVE_INVENTORY.equals(order.getOrderStatus())) { productService.handleOrder(order) .doOnSuccess(o -> { orderProducer.sendMessage(order.setOrderStatus(OrderStatus.INVENTORY_SUCCESS)); }) .doOnError(e -> { orderProducer.sendMessage(order.setOrderStatus(OrderStatus.INVENTORY_FAILURE) .setResponseMessage(e.getMessage())); }).subscribe(); } else if (OrderStatus.REVERT_INVENTORY.equals(order.getOrderStatus())) { productService.revertOrder(order) .doOnSuccess(o -> { orderProducer.sendMessage(order.setOrderStatus(OrderStatus.INVENTORY_REVERT_SUCCESS)); }) .doOnError(e -> { orderProducer.sendMessage(order.setOrderStatus(OrderStatus.INVENTORY_REVERT_FAILURE) .setResponseMessage(e.getMessage())); }).subscribe(); } }

This also means that we can safely drop some of the redundant endpoints from our controller now. These changes are sufficient to achieve asynchronous communication in our application.

6.2. Shipping Service

The changes in shipping service are relatively similar to what we did earlier with the inventory service. The message producer is the same, and the message consumer is specific to shipping logic:

@KafkaListener(topics = "orders", groupId = "shipping") public void consume(Order order) throws IOException { if (OrderStatus.PREPARE_SHIPPING.equals(order.getOrderStatus())) { shippingService.handleOrder(order) .doOnSuccess(o -> { orderProducer.sendMessage(order.setOrderStatus(OrderStatus.SHIPPING_SUCCESS) .setShippingDate(o.getShippingDate())); }) .doOnError(e -> { orderProducer.sendMessage(order.setOrderStatus(OrderStatus.SHIPPING_FAILURE) .setResponseMessage(e.getMessage())); }).subscribe(); } }

We can safely drop all the endpoints in our controller now as we no longer need them.

6.3. Order Service

The changes in order service will be a little more involved as this is where we were doing all the orchestration earlier.

Nevertheless, the message producer remains unchanged, and message consumer takes on order service-specific logic:

@KafkaListener(topics = "orders", groupId = "orders") public void consume(Order order) throws IOException { if (OrderStatus.INITIATION_SUCCESS.equals(order.getOrderStatus())) { orderRepository.findById(order.getId()) .map(o -> { orderProducer.sendMessage(o.setOrderStatus(OrderStatus.RESERVE_INVENTORY)); return o.setOrderStatus(order.getOrderStatus()) .setResponseMessage(order.getResponseMessage()); }) .flatMap(orderRepository::save) .subscribe(); } else if ("INVENTORY-SUCCESS".equals(order.getOrderStatus())) { orderRepository.findById(order.getId()) .map(o -> { orderProducer.sendMessage(o.setOrderStatus(OrderStatus.PREPARE_SHIPPING)); return o.setOrderStatus(order.getOrderStatus()) .setResponseMessage(order.getResponseMessage()); }) .flatMap(orderRepository::save) .subscribe(); } else if ("SHIPPING-FAILURE".equals(order.getOrderStatus())) { orderRepository.findById(order.getId()) .map(o -> { orderProducer.sendMessage(o.setOrderStatus(OrderStatus.REVERT_INVENTORY)); return o.setOrderStatus(order.getOrderStatus()) .setResponseMessage(order.getResponseMessage()); }) .flatMap(orderRepository::save) .subscribe(); } else { orderRepository.findById(order.getId()) .map(o -> { return o.setOrderStatus(order.getOrderStatus()) .setResponseMessage(order.getResponseMessage()); }) .flatMap(orderRepository::save) .subscribe(); } }

The consumer here is merely reacting to order messages with different order statuses. This is what gives us the choreography between different services.

Lastly, our order service will also have to change to support this choreography:

public Mono createOrder(Order order) { return Mono.just(order) .flatMap(orderRepository::save) .map(o -> { orderProducer.sendMessage(o.setOrderStatus(OrderStatus.INITIATION_SUCCESS)); return o; }) .onErrorResume(err -> { return Mono.just(order.setOrderStatus(OrderStatus.FAILURE) .setResponseMessage(err.getMessage())); }) .flatMap(orderRepository::save); }

Note that this is far simpler than the service we had to write with reactive endpoints in the last section. Asynchronous choreography often results in far simpler code, although it does come at the cost of eventual consistency and complex debugging and monitoring. As we may guess, our front-end will no longer get the final status of the order immediately.

7. Container Orchestration Service

The last piece of the puzzle that we want to solve is related to deployment.

What we want in the application is ample redundancy and a tendency to scale up or down depending upon the need automatically.

We've already achieved containerization of services through Docker and are managing dependencies between them through Docker Compose. While these are fantastic tools in their own right, they do not help us to achieve what we want.

Hence, we need a container orchestration service that can take care of redundancy and scalability in our application. While there are several options, one of the popular ones includes Kubernetes. Kubernetes provides us with a cloud vendor-agnostic way to achieve highly scalable deployments of containerized workloads.

Kubernetes wraps containers like Docker into Pods, which are the smallest unit of deployment. Further, we can use Deployment to describe the desired state declaratively.

Deployment creates ReplicaSets, which internally is responsible for bringing up the pods. We can describe a minimum number of identical pods that should be running at any point in time. This provides redundancy and hence high availability.

Let's see how can we define a Kubernetes deployment for our applications:

apiVersion: apps/v1 kind: Deployment metadata: name: inventory-deployment spec: replicas: 3 selector: matchLabels: name: inventory-deployment template: metadata: labels: name: inventory-deployment spec: containers: - name: inventory image: inventory-service-async:latest ports: - containerPort: 8081 --- apiVersion: apps/v1 kind: Deployment metadata: name: shipping-deployment spec: replicas: 3 selector: matchLabels: name: shipping-deployment template: metadata: labels: name: shipping-deployment spec: containers: - name: shipping image: shipping-service-async:latest ports: - containerPort: 8082 --- apiVersion: apps/v1 kind: Deployment metadata: name: order-deployment spec: replicas: 3 selector: matchLabels: name: order-deployment template: metadata: labels: name: order-deployment spec: containers: - name: order image: order-service-async:latest ports: - containerPort: 8080

Here we're declaring our deployment to maintain three identical replicas of pods at any time. While this is a good way to add redundancy, it may not be sufficient for varying loads. Kubernetes provides another resource known as the Horizontal Pod Autoscaler which can scale the number of pods in a deployment based on observed metrics like CPU utilization.

Please note that we have just covered the scalability aspects of the application hosted on a Kubernetes cluster. This does not necessarily imply that the underlying cluster itself is scalable. Creating a high availability Kubernetes cluster is a non-trivial task and beyond the scope of this tutorial.

8. Resulting Reactive System

Now that we've made several improvements in our architecture, it's perhaps time to evaluate this against the definition of a Reactive System. We'll keep the evaluation against the four characteristics of a Reactive Systems we discussed earlier in the tutorial:

  • Responsive: The adoption of the reactive programming paradigm should help us achieve end-to-end non-blocking and hence a responsive application
  • Resilient: Kubernetes deployment with ReplicaSet of the desired number of pods should provide resilience against random failures
  • Elastic: Kubernetes cluster and resources should provide us the necessary support to be elastic in the face of unpredictable loads
  • Message-Driven: Having all service-to-service communication handled asynchronously through a Kafka broker should help us here

While this looks quite promising, it's far from over. To be honest, the quest for a truly reactive system should be a continuous exercise of improvements. We can never preempt all that can fail in a highly complex infrastructure, where our application is just a small part.

A reactive system thus will demand reliability from every part that makes the whole. Right from the physical network to infrastructure services like DNS, they all should fall in line to help us achieve the end goal.

Often, it may not be possible for us to manage and provide the necessary guarantees for all these parts. And this is where a managed cloud infrastructure helps alleviate our pain. We can choose from a host of services like IaaS (Infeastrure-as-a-Service), BaaS (Backend-as-a-Service), and PaaS (Platform-as-a-Service) to delegate the responsibilities to external parties. This leaves us with the responsibility of our application as far as possible.

9. Conclusion

In this tutorial, we went through the basics of reactive systems and how does it compare with reactive programming. We created a simple application with multiple microservices and highlighted the problems we intend to solve with a reactive system.

Inoltre, siamo andati avanti, introducendo la programmazione reattiva, l'architettura basata sui messaggi e il servizio di orchestrazione dei contenitori nell'architettura per realizzare un sistema reattivo.

Infine, abbiamo discusso dell'architettura risultante e di come rimane un viaggio verso il sistema reattivo! Questo tutorial non ci introduce a tutti gli strumenti, framework o pattern che possono aiutarci a creare un sistema reattivo, ma ci introduce al viaggio.

Come al solito, il codice sorgente di questo articolo può essere trovato su GitHub.