Example of mixing stream with completable future using composition.

Please refer my other github project: https://github.com/mtumilowicz/java11-stream-completablefuture-dedicated-executor about providing dedicated thread pool for completable future tasks.

Suppose we have stream of product ids and we want to perform operations:

1. get the product by its id
2. pack the product
3. send the product
4. generate the report

Note that operation could be time consuming, for example getting product by its id could be done by querying external service.

• Delay

  class Delay {
      static void delay() {
          try {
              Thread.sleep(600);
          } catch (InterruptedException e) {
              // not used
          }
      }
  }
  

• classes Product, Packed and Send are as simple as they can be (with some utility functions):

  @Value
  class Product {
      int id;
      
      static Function<Integer, CompletableFuture<Product>> getProduct(Function<Integer, Product> productProvider,
                                                                      Executor executor) {
          return id -> CompletableFuture.supplyAsync(() -> productProvider.apply(id), executor);
      }
  }
  
  @Value
  class Packed {
      Product product;
  
      static Function<Product, CompletableFuture<Packed>> pack(Executor executor) {
          return product -> CompletableFuture.supplyAsync(() -> new Packed(product), executor);
      }
  }
  
  @Value
  class Send {
      Packed packed;
      
      static Function<Packed, CompletableFuture<Send>> send(Executor executor) {
          return packed -> CompletableFuture.supplyAsync(() -> new Send(packed), executor);
      }
      
      String asReport() {
          return "Successfully send: " + packed.toString();
      }
  }
  

• ProductService

  • packing (two approaches)

    • sequence of maps

      List<String> send(List<Integer> ids) {
      
          var executors = productThreadPool(ids.size());
      
          var sendFutures = ids.stream()
                  .map(id -> CompletableFuture.supplyAsync(() -> new Product(id), executors))
                  .map(product -> product.thenCompose(Packed.pack(by(executors))))
                  .map(packed -> packed.thenCompose(Send.send(by(executors))))
                  .map(send -> send.thenApply(Send::toString))
                  .map(future -> future.orTimeout(500, TimeUnit.MILLISECONDS))
                  .map(future -> future.handle((ok, ex) -> nonNull(ok) ? ok : "FAILED: " + ex))
                  .collect(toList());
      
          return sendFutures.stream()
                  .map(CompletableFuture::join)
                  .collect(toList());
      }
      

      cons: weak error handling - we cannot customize message in a reasonable way - for example we do not have access to the product id.
    • map with composed CompletableFuture

      var executors = productThreadPool(ids.size());
      
      var sendFutures = ids.stream()
              .map(id -> Product.getProduct(productsProvider).apply(id)
                      .thenCompose(Packed.pack(by(executors)))
                      .thenCompose(Send.send(by(executors)))
                      .thenApply(Send::asReport)
                      .orTimeout(500, TimeUnit.MILLISECONDS)
                      .handle((ok, ex) -> nonNull(ok) ? ok : "FAILED with product = " + id + ": " + ex))
              .collect(toList());
      
      return sendFutures.stream()
              .map(CompletableFuture::join)
              .collect(toList());
      

      pros: we have direct access to id when it comes to error handling

    we want to stress that it is nearly always a good idea to support timeout and handle exception when using completable future:

    • future -> future.orTimeout(500, TimeUnit.MILLISECONDS)
    • future -> future.handle((ok, ex) -> nonNull(ok) ? ok : "FAILED: " + ex)

  • executors

    private Executor productThreadPool(int size) {
        return Executors.newFixedThreadPool(Math.min(size, 100),
                r -> {
                    Thread t = new Thread(r);
                    t.setDaemon(true);
                    return t;
                });
    }
    

  • and small utility function for readability:

    private Executor by(Executor executor) {
        return executor;
    }