AsyncIterablePublisher.java

/***************************************************
 * Licensed under MIT No Attribution (SPDX: MIT-0) *
 ***************************************************/

package org.reactivestreams.example.unicast;

import org.reactivestreams.Publisher;
import org.reactivestreams.Subscriber;
import org.reactivestreams.Subscription;

import java.util.Iterator;
import java.util.Collections;
import java.util.concurrent.Executor;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.ConcurrentLinkedQueue;

/**
 * AsyncIterablePublisher is an implementation of Reactive Streams `Publisher`
 * which executes asynchronously, using a provided `Executor` and produces elements
 * from a given `Iterable` in a "unicast" configuration to its `Subscribers`.
 *
 * NOTE: The code below uses a lot of try-catches to show the reader where exceptions can be expected, and where they are forbidden.
 */
public class AsyncIterablePublisher<T> implements Publisher<T> {
  private final static int DEFAULT_BATCHSIZE = 1024;

  private final Iterable<T> elements; // This is our data source / generator
  private final Executor executor; // This is our thread pool, which will make sure that our Publisher runs asynchronously to its Subscribers
  private final int batchSize; // In general, if one uses an `Executor`, one should be nice nad not hog a thread for too long, this is the cap for that, in elements

  public AsyncIterablePublisher(final Iterable<T> elements, final Executor executor) {
    this(elements, DEFAULT_BATCHSIZE, executor);
  }

  public AsyncIterablePublisher(final Iterable<T> elements, final int batchSize, final Executor executor) {
    if (elements == null) throw null;
    if (executor == null) throw null;
    if (batchSize < 1) throw new IllegalArgumentException("batchSize must be greater than zero!");
    this.elements = elements;
    this.executor = executor;
    this.batchSize = batchSize;
  }

  @Override
  public void subscribe(final Subscriber<? super T> s) {
    // As per rule 1.11, we have decided to support multiple subscribers in a unicast configuration
    // for this `Publisher` implementation.
    // As per 2.13, this method must return normally (i.e. not throw)
    new SubscriptionImpl(s).init();
  }

  // These represent the protocol of the `AsyncIterablePublishers` SubscriptionImpls
  static interface Signal {};
  enum Cancel implements Signal { Instance; };
  enum Subscribe implements Signal { Instance; };
  enum Send implements Signal { Instance; };
  static final class Request implements Signal {
    final long n;
    Request(final long n) {
      this.n = n;
    }
  };

  // This is our implementation of the Reactive Streams `Subscription`,
  // which represents the association between a `Publisher` and a `Subscriber`.
  final class SubscriptionImpl implements Subscription, Runnable {
    final Subscriber<? super T> subscriber; // We need a reference to the `Subscriber` so we can talk to it
    private boolean cancelled = false; // This flag will track whether this `Subscription` is to be considered cancelled or not
    private long demand = 0; // Here we track the current demand, i.e. what has been requested but not yet delivered
    private Iterator<T> iterator; // This is our cursor into the data stream, which we will send to the `Subscriber`

    SubscriptionImpl(final Subscriber<? super T> subscriber) {
      // As per rule 1.09, we need to throw a `java.lang.NullPointerException` if the `Subscriber` is `null`
      if (subscriber == null) throw null;
      this.subscriber = subscriber;
    }

    // This `ConcurrentLinkedQueue` will track signals that are sent to this `Subscription`, like `request` and `cancel`
    private final ConcurrentLinkedQueue<Signal> inboundSignals = new ConcurrentLinkedQueue<Signal>();

    // We are using this `AtomicBoolean` to make sure that this `Subscription` doesn't run concurrently with itself,
    // which would violate rule 1.3 among others (no concurrent notifications).
    private final AtomicBoolean on = new AtomicBoolean(false);

    // This method will register inbound demand from our `Subscriber` and validate it against rule 3.9 and rule 3.17
    private void doRequest(final long n) {
      if (n < 1)
        terminateDueTo(new IllegalArgumentException(subscriber + " violated the Reactive Streams rule 3.9 by requesting a non-positive number of elements."));
      else if (demand + n < 1) {
        // As governed by rule 3.17, when demand overflows `Long.MAX_VALUE` we treat the signalled demand as "effectively unbounded"
        demand = Long.MAX_VALUE;  // Here we protect from the overflow and treat it as "effectively unbounded"
        doSend(); // Then we proceed with sending data downstream
      } else {
        demand += n; // Here we record the downstream demand
        doSend(); // Then we can proceed with sending data downstream
      }
    }

    // This handles cancellation requests, and is idempotent, thread-safe and not synchronously performing heavy computations as specified in rule 3.5
    private void doCancel() {
      cancelled = true;
    }

    // Instead of executing `subscriber.onSubscribe` synchronously from within `Publisher.subscribe`
    // we execute it asynchronously, this is to avoid executing the user code (`Iterable.iterator`) on the calling thread.
    // It also makes it easier to follow rule 1.9
    private void doSubscribe() {
      try {
        iterator = elements.iterator();
        if (iterator == null)
          iterator = Collections.<T>emptyList().iterator(); // So we can assume that `iterator` is never null
      } catch(final Throwable t) {
        subscriber.onSubscribe(new Subscription() { // We need to make sure we signal onSubscribe before onError, obeying rule 1.9
          @Override public void cancel() {}
          @Override public void request(long n) {}
        });
        terminateDueTo(t); // Here we send onError, obeying rule 1.09
      }

      if (!cancelled) {
        // Deal with setting up the subscription with the subscriber
        try {
          subscriber.onSubscribe(this);
        } catch(final Throwable t) { // Due diligence to obey 2.13
          terminateDueTo(new IllegalStateException(subscriber + " violated the Reactive Streams rule 2.13 by throwing an exception from onSubscribe.", t));
        }

        // Deal with already complete iterators promptly
        boolean hasElements = false;
        try {
          hasElements = iterator.hasNext();
        } catch(final Throwable t) {
          terminateDueTo(t); // If hasNext throws, there's something wrong and we need to signal onError as per 1.2, 1.4, 
        }

        // If we don't have anything to deliver, we're already done, so lets do the right thing and
        // not wait for demand to deliver `onComplete` as per rule 1.2 and 1.3
        if (!hasElements) {
          try {
            doCancel(); // Rule 1.6 says we need to consider the `Subscription` cancelled when `onComplete` is signalled
            subscriber.onComplete();
          } catch(final Throwable t) { // As per rule 2.13, `onComplete` is not allowed to throw exceptions, so we do what we can, and log this.
            (new IllegalStateException(subscriber + " violated the Reactive Streams rule 2.13 by throwing an exception from onComplete.", t)).printStackTrace(System.err);
          }
        }
      }
    }

    // This is our behavior for producing elements downstream
    private void doSend() {
      try {
        // In order to play nice with the `Executor` we will only send at-most `batchSize` before
        // rescheduing ourselves and relinquishing the current thread.
        int leftInBatch = batchSize;
        do {
          T next;
          boolean hasNext;
          try {
            next = iterator.next(); // We have already checked `hasNext` when subscribing, so we can fall back to testing -after- `next` is called.
            hasNext = iterator.hasNext(); // Need to keep track of End-of-Stream
          } catch (final Throwable t) {
            terminateDueTo(t); // If `next` or `hasNext` throws (they can, since it is user-provided), we need to treat the stream as errored as per rule 1.4
            return;
          }
          subscriber.onNext(next); // Then we signal the next element downstream to the `Subscriber`
          if (!hasNext) { // If we are at End-of-Stream
            doCancel(); // We need to consider this `Subscription` as cancelled as per rule 1.6
            subscriber.onComplete(); // Then we signal `onComplete` as per rule 1.2 and 1.5
          }
        } while (!cancelled           // This makes sure that rule 1.8 is upheld, i.e. we need to stop signalling "eventually"
                 && --leftInBatch > 0 // This makes sure that we only send `batchSize` number of elements in one go (so we can yield to other Runnables)
                 && --demand > 0);    // This makes sure that rule 1.1 is upheld (sending more than was demanded)

        if (!cancelled && demand > 0) // If the `Subscription` is still alive and well, and we have demand to satisfy, we signal ourselves to send more data
          signal(Send.Instance);
      } catch(final Throwable t) {
        // We can only get here if `onNext` or `onComplete` threw, and they are not allowed to according to 2.13, so we can only cancel and log here.
        doCancel(); // Make sure that we are cancelled, since we cannot do anything else since the `Subscriber` is faulty.
        (new IllegalStateException(subscriber + " violated the Reactive Streams rule 2.13 by throwing an exception from onNext or onComplete.", t)).printStackTrace(System.err);
      }
    }

    // This is a helper method to ensure that we always `cancel` when we signal `onError` as per rule 1.6
    private void terminateDueTo(final Throwable t) {
      cancelled = true; // When we signal onError, the subscription must be considered as cancelled, as per rule 1.6
      try {
        subscriber.onError(t); // Then we signal the error downstream, to the `Subscriber`
      } catch(final Throwable t2) { // If `onError` throws an exception, this is a spec violation according to rule 1.9, and all we can do is to log it.
        (new IllegalStateException(subscriber + " violated the Reactive Streams rule 2.13 by throwing an exception from onError.", t2)).printStackTrace(System.err);
      }
    }

    // What `signal` does is that it sends signals to the `Subscription` asynchronously
    private void signal(final Signal signal) {
      if (inboundSignals.offer(signal)) // No need to null-check here as ConcurrentLinkedQueue does this for us
        tryScheduleToExecute(); // Then we try to schedule it for execution, if it isn't already
    }

    // This is the main "event loop" if you so will
    @Override public final void run() {
      if(on.get()) { // establishes a happens-before relationship with the end of the previous run
        try {
          final Signal s = inboundSignals.poll(); // We take a signal off the queue
          if (!cancelled) { // to make sure that we follow rule 1.8, 3.6 and 3.7

            // Below we simply unpack the `Signal`s and invoke the corresponding methods
            if (s instanceof Request)
              doRequest(((Request)s).n);
            else if (s == Send.Instance)
              doSend();
            else if (s == Cancel.Instance)
              doCancel();
            else if (s == Subscribe.Instance)
              doSubscribe();
          }
        } finally {
          on.set(false); // establishes a happens-before relationship with the beginning of the next run
          if(!inboundSignals.isEmpty()) // If we still have signals to process
            tryScheduleToExecute(); // Then we try to schedule ourselves to execute again
        }
      }
    }

    // This method makes sure that this `Subscription` is only running on one Thread at a time,
    // this is important to make sure that we follow rule 1.3
    private final void tryScheduleToExecute() {
      if(on.compareAndSet(false, true)) {
        try {
          executor.execute(this);
        } catch(Throwable t) { // If we can't run on the `Executor`, we need to fail gracefully
          if (!cancelled) {
            doCancel(); // First of all, this failure is not recoverable, so we need to follow rule 1.4 and 1.6
            try {
              terminateDueTo(new IllegalStateException("Publisher terminated due to unavailable Executor.", t));
            } finally {
              inboundSignals.clear(); // We're not going to need these anymore
              // This subscription is cancelled by now, but letting it become schedulable again means
              // that we can drain the inboundSignals queue if anything arrives after clearing
              on.set(false);
            }
          }
        }
      }
    }

    // Our implementation of `Subscription.request` sends a signal to the Subscription that more elements are in demand
    @Override public void request(final long n) {
      signal(new Request(n));
    }
    // Our implementation of `Subscription.cancel` sends a signal to the Subscription that the `Subscriber` is not interested in any more elements
    @Override public void cancel() {
      signal(Cancel.Instance);
    }
    // The reason for the `init` method is that we want to ensure the `SubscriptionImpl`
    // is completely constructed before it is exposed to the thread pool, therefor this
    // method is only intended to be invoked once, and immediately after the constructor has
    // finished.
    void init() {
      signal(Subscribe.Instance);
    }
  };
}