在 RxJava 服务中管理事务性的正确方法是什么？

Question

我最近开始尝试使用 RxJava 并看到一位 Netflix 工程师的演示文稿，该演示文稿建议将我们的业务 API 迁移到 Observable API，例如：

public interface VideoService {
    Observable<VideoBasicInfo> createVideoBasicInfo(VideoBasicInfo videoBasic);
    Observable<VideoBasicInfo> getVideoBasicInfo(Integer videoId);
    Observable<VideoRating> getVideoRating(Integer videoId);
}

但是我还没有找到任何地方来解释应该如何在此服务中管理事务性。起初我只是用 @Transactional

注释我的服务实现

@Service
@Transactional
public class VideoServiceImpl implements VideoService{

    @Autowired
    private VideoBasicInfoRepository basicInfoRepo;
    @Autowired
    private VideoRatingRepository ratingRepo;

    public Observable<VideoBasicInfo> createVideoBasicInfo(VideoBasicInfo videoBasic){
        return Observable.create( s -> {
            s.onNext(basicInfoRepo.save(videBasic));
        });
    }

我们想要的是 Object.create lambda (s -> { // This code }) 中所有代码的执行发生在一个事务中。然而，实际发生的是：

对 createVideoBasicInfo() 的调用以事务方式执行，返回冷 observable。
save() 作为原子事务执行。

显然这是有道理的，因为 Spring 代理适用于 serviceImpl 方法。我已经想到了一些方法来做我真正期望的事情，例如开始程序化交易：

return Observable.create( s -> {
    VideoBasicInfo savedBasic = transactionTemplate.execute( status -> {
        VideoBasicInfo basicInfo = basicInfoRepo.save(videoBasicInfo);
        return basicInfo;
    });
    s.onNext(savedBasic);
});

这是使用响应式 API 时管理事务的推荐方式吗？

Answer 1

Spring Data JpaRepository 方法签名已经被标记为@Transactional，所以如果你只使用一个那么你不需要做任何特殊的事情：

public interface PersonRepository extends JpaRepository<Person, Integer> {
}

@RunWith(SpringJUnit4ClassRunner.class)
@SpringApplicationConfiguration(classes = {RepositoryConfiguration.class})
public class PersonRepositoryTest {
    private PersonRepository personRepository;

    @Autowired
    public void setPersonRepository(PersonRepository PersonRepository) {
        this.personRepository = PersonRepository;
    }

    @Test
    public void testReactiveSavePerson() {
        Person person = new Person("Jane", "Doe");
        assertNull(person.getId()); //null before save

        //save person
        Observable.create(s -> {
            s.onNext(personRepository.save(person));
        }).subscribe();

        //fetch from DB
        Person fetchedPerson = personRepository.findOne(person.getId());

        //should not be null
        assertNotNull(fetchedPerson);

        //should equal
        assertEquals(person.getId(), fetchedPerson.getId());
        assertEquals(person.getFirstName(), fetchedPerson.getFirstName());
    }
}

如果您需要将多个存储库合并到一个事务中，您可以使用类似下面的 class 的东西：

@Component()
public class ObservableTxFactory {
    public final <T> Observable<T> create(Observable.OnSubscribe<T> f) {
        return new ObservableTx<>(this, f);
    }

    @Transactional
    public void call(Observable.OnSubscribe onSubscribe, Subscriber subscriber) {
        onSubscribe.call(subscriber);
    }

    private static class ObservableTx<T> extends Observable<T> {

        public ObservableTx(ObservableTxFactory observableTxFactory, OnSubscribe<T> f) {
            super(new OnSubscribeDecorator<>(observableTxFactory, f));
        }
    }

    private static class OnSubscribeDecorator<T> implements Observable.OnSubscribe<T> {

        private final ObservableTxFactory observableTxFactory;
        private final Observable.OnSubscribe<T> onSubscribe;

        OnSubscribeDecorator(final ObservableTxFactory observableTxFactory, final Observable.OnSubscribe<T> s) {
            this.onSubscribe = s;
            this.observableTxFactory = observableTxFactory;
        }

        @Override
        public void call(Subscriber<? super T> subscriber) {
            observableTxFactory.call(onSubscribe, subscriber);
        }
    }
}

工厂bean也需要定义：

@Bean
ObservableTxFactory observableTxFactory() {
    return new ObservableTxFactory();
}

服务：

@Service
public class PersonService {
    @Autowired
    PersonRepository personRepository;
    @Autowired
    ObservableTxFactory observableTxFactory;

    public Observable<Person> createPerson(String firstName, String lastName) {
        return observableTxFactory.create(s -> {
            Person p = new Person(firstName, lastName);
            s.onNext(personRepository.save(p));
        });
    }
}

测试：

@RunWith(SpringJUnit4ClassRunner.class)
@SpringApplicationConfiguration(classes = {RepositoryConfiguration.class})
public class PersonServiceTest {
    @Autowired
    PersonRepository personRepository;
    @Autowired
    ObservableTxFactory observableTxFactory;

    @Test
    public void testPersonService() {
        final PersonService service = new PersonService();
        service.personRepository = personRepository;
        service.observableTxFactory = observableTxFactory;

        final Observable<Person> personObservable = service.createPerson("John", "Doe");
        personObservable.subscribe();

        //fetch from DB
        final Person fetchedPerson = StreamSupport.stream(personRepository.findAll().spliterator(), false)
                .filter(p -> p.getFirstName().equals("John") && p.getLastName().equals("Doe"))
                .findFirst()
                .get();

        //should not be null
        assertNotNull(fetchedPerson);
    }

}

显示代理的屏幕截图：

Answer 2

我想借用 John Scattergood 的出色回答。我的典型用法是使用 Observable.fromCallable() 所以我正在寻找一种方法来做到这一点而不是实现 Observable.OnSubscribe 所以我调整了他的技术以便您可以通过传入 Callable 来使用它

工厂Class:

@Component
public class ObservableTxFactory {
    public final <T> Observable.OnSubscribe<T> createFromCallable(Callable<? extends T> resultFactory) {
        return new OnSubscribeDecorator<>(this, resultFactory);
    }

    @SuppressWarnings("unchecked")
    @Transactional
    public <T> void call(Callable<? extends T> resultFactory, Subscriber subscriber) {
        final SingleDelayedProducer<T> singleDelayedProducer = new SingleDelayedProducer<>(subscriber);

        subscriber.setProducer(singleDelayedProducer);

        try {
            singleDelayedProducer.setValue(resultFactory.call());
        } catch (Throwable t) {
            Exceptions.throwOrReport(t, subscriber);
        }
    }

    private static class OnSubscribeDecorator<T> implements Observable.OnSubscribe<T> {

        private final ObservableTxFactory observableTxFactory;
        private final Callable<? extends T> resultFactory;

        OnSubscribeDecorator(final ObservableTxFactory observableTxFactory, Callable<? extends T> resultFactory) {
            this.resultFactory = resultFactory;
            this.observableTxFactory = observableTxFactory;
        }

        @Override
        public void call(Subscriber<? super T> subscriber) {
            observableTxFactory.call(resultFactory, subscriber);
        }
    }
}

原码：

Observable.fromCallable(() -> fooRepository.findOne(fooID));

新代码：

Observable.create(observableTxFactory.createFromCallable(() -> fooRepository.findOne(fooID)));

确保你添加 @Transactional 的方法是 public 否则 Spring AOP 将无法建议它

在 RxJava 服务中管理事务性的正确方法是什么？

What is the correct way of managing transactionality in RxJava Services?

spring

netflix

reactive-programming

rx-java

rx-android