使用Spring MVC流可关闭资源

After having read this article, I wish to use Spring to stream database query results directly to a JSON response to ensure constant-memory usage (no greedy loading of a List in memory).

阅读完本文之后，我希望使用Spring将数据库查询结果直接流式传输到JSON响应，以确保内存使用率不变（内存中没有贪婪的List加载）。

Similar to what is done in the article with Hibernate, I assembled a greetingRepository object which returns a stream of the database contents based on a JdbcTemplate. In that implementation, I create an iterator over the queried ResultSet, and I return the stream as follows:

与使用Hibernate的文章中所做的类似，我组装了一个greetingRepository对象，该对象根据JdbcTemplate返回数据库内容的流。在该实现中，我在查询的ResultSet上创建了一个迭代器，并按如下方式返回流：

return StreamSupport.stream(spliterator(), false).onClose(() -> {
  log.info("Closing ResultSetIterator stream");
  JdbcUtils.closeResultSet(resultSet);
});

i.e. with an onClose() method guaranteeing that the underlying ResultSet will be closed if the stream is declared in a try-with-resources construct:

即使用onClose（）方法，如果在try-with-resources构造中声明了流，则保证将关闭基础ResultSet：

try(Stream<Greeting> stream = greetingRepository.stream()) {
  // operate on the stream
} // ResultSet underlying the stream will be guaranteed to be closed

But as in the article, I want this stream to be consumed by a custom object mapper (the enhanced MappingJackson2HttpMessageConverter defined in the article). If we take the try-with-resources need aside, this is feasible as follows:

但正如在文章中，我希望这个流由自定义对象映射器（文章中定义的增强的MappingJackson2HttpMessageConverter）使用。如果我们将资源需求放在一边，这可行如下：

@RequestMapping(method = GET)
Stream<GreetingResource> stream() {
  return greetingRepository.stream().map(GreetingResource::new);
}

However as a colleague commented at the bottom of that article, this does not take care of closing underlying resources.

然而，正如一位同事在该文章底部评论的那样，这并没有关闭基础资源。

In the context of Spring MVC, how can I stream from the database all the way into a JSON response and still guarantee that the ResultSet will be closed? Could you provide a concrete example solution?

在Spring MVC的上下文中，我如何从数据库一直流式传输到JSON响应中并仍然保证ResultSet将被关闭？你能提供具体的示例解决方案吗？

1 个解决方案

#1

You could create a construct to defer the query execution at the serialization time. This construct will start and end the transaction programmaticaly.

您可以创建一个构造以在序列化时延迟查询执行。此构造将以programmaticaly方式开始和结束事务。

public class TransactionalStreamable<T> {

    private final PlatformTransactionManager platformTransactionManager;

    private final Callable<Stream<T>> callable;

    public TransactionalStreamable(PlatformTransactionManager platformTransactionManager, Callable<Stream<T>> callable) {
        this.platformTransactionManager = platformTransactionManager;
        this.callable = callable;
    }

    public Stream stream() {
        TransactionTemplate txTemplate = new TransactionTemplate(platformTransactionManager);
        txTemplate.setPropagationBehavior(TransactionDefinition.PROPAGATION_REQUIRES_NEW);
        txTemplate.setReadOnly(true);

        TransactionStatus transaction = platformTransactionManager.getTransaction(txTemplate);

        try {
            return callable.call().onClose(() -> {
                platformTransactionManager.commit(transaction);
            });
        } catch (Exception e) {
            platformTransactionManager.rollback(transaction);
            throw new RuntimeException(e);
        }
    }

    public void forEach(Consumer<T> c) {
        try (Stream<T> s = stream()){
            s.forEach(c);
        }
    }
}

Using a dedicated json serializer:

使用专用的json序列化器：

JsonSerializer<?> transactionalStreamableSer = new StdSerializer<TransactionalStreamable<?>>(TransactionalStreamable.class, true) {
    @Override
    public void serialize(TransactionalStreamable<?> streamable, JsonGenerator jgen, SerializerProvider provider) throws IOException {
        jgen.writeStartArray();
        streamable.forEach((CheckedConsumer) e -> {
            provider.findValueSerializer(e.getClass(), null).serialize(e, jgen, provider);
        });

        jgen.writeEndArray();
    }
};

Which could be used like this:

可以这样使用：

@RequestMapping(method = GET)
TransactionalStreamable<GreetingResource> stream() {
  return new TransactionalStreamable(platformTransactionManager , () -> greetingRepository.stream().map(GreetingResource::new));
}

All the work will be done when jackson will serialize the object. It may be or not an issue regarding the error handling (eg. using controller advice).

当杰克逊将序列化对象时，所有工作都将完成。它可能是或者不是关于错误处理的问题（例如，使用控制器建议）。

#1