我想要达到的目的是根据用户对我从Kafka主题接收到的一些消息进行分组，并打开它们的窗口，以便在（5分钟）窗口中聚合我接收到的消息。然后我想收集每个窗口中的所有聚合，以便立即处理它们，并将它们添加到我在5分钟间隔内收到的所有消息的报告中。
最后一点似乎是困难的部分，因为Kafka流似乎没有提供（至少我找不到它！）任何可以在一个“有限”流中收集所有与窗口相关的内容并在一个地方进行处理的东西。
这是我实现的代码

StreamsBuilder builder = new StreamsBuilder();
KStream<UserId, Message> messages = builder.stream("KAFKA_TOPIC");

TimeWindowedKStream<UserId, Message> windowedMessages =
        messages.
                groupByKey().windowedBy(TimeWindows.of(SIZE_MS));

KTable<Windowed<UserId>, List<Message>> messagesAggregatedByWindow =
        windowedMessages.
                aggregate(
                        () -> new LinkedList<>(), new MyAggregator<>(),
                        Materialized.with(new MessageKeySerde(), new MessageListSerde())
                );

messagesAggregatedByWindow.toStream().foreach((key, value) -> log.info("({}), KEY {} MESSAGE {}",  value.size(), key, value.toString()));

KafkaStreams streams = new KafkaStreams(builder.build(), config);
streams.start();

结果是

KEY [UserId(82770583)@1531502760000/1531502770000] Message [Message(userId=UserId(82770583),message="a"),Message(userId=UserId(82770583),message="b"),Message(userId=UserId(82770583),message="d")]
KEY [UserId(77082590)@1531502760000/1531502770000] Message [Message(userId=UserId(77082590),message="g")]
KEY [UserId(85077691)@1531502750000/1531502760000] Message [Message(userId=UserId(85077691),message="h")]
KEY [UserId(79117307)@1531502780000/1531502790000] Message [Message(userId=UserId(79117307),message="e")]
KEY [UserId(73176289)@1531502760000/1531502770000] Message [Message(userId=UserId(73176289),message="r"),Message(userId=UserId(73176289),message="q")]
KEY [UserId(92077080)@1531502760000/1531502770000] Message [Message(userId=UserId(92077080),message="w")]
KEY [UserId(78530050)@1531502760000/1531502770000] Message [Message(userId=UserId(78530050),message="t")]
KEY [UserId(64640536)@1531502760000/1531502770000] Message [Message(userId=UserId(64640536),message="y")]

每个窗口都有许多日志行，它们与其他窗口混合在一起。
我想要的是：

// Hypothetical implementation
windowedMessages.streamWindows((interval, window) -> process(interval, window));

方法过程类似于：

// Hypothetical implementation

void process(Interval interval, WindowStream<UserId, List<Message>> windowStream) {
// Create report for the whole window   
Report report = new Report(nameFromInterval());
    // Loop on the finite iterable that represents the window content
    for (WindowStreamEntry<UserId, List<Message>> entry: windowStream) {
        report.addLine(entry.getKey(), entry.getValue());
    }
    report.close();
}

结果将按如下方式分组（每个报告都是对my callback:void process（…）的调用），每个窗口的提交将在处理整个窗口时提交：

Report 1:
    KEY [UserId(85077691)@1531502750000/1531502760000] Message [Message(userId=UserId(85077691),message="h")]

Report 2:
    KEY [UserId(82770583)@1531502760000/1531502770000] Message [Message(userId=UserId(82770583),message="a"),Message(userId=UserId(82770583),message="b"),Message(userId=UserId(82770583),message="d")]
    KEY [UserId(77082590)@1531502760000/1531502770000] Message [Message(userId=UserId(77082590),message="g")]
    KEY [UserId(73176289)@1531502760000/1531502770000] Message [Message(userId=UserId(73176289),message="r"),Message(userId=UserId(73176289),message="q")]
    KEY [UserId(92077080)@1531502760000/1531502770000] Message [Message(userId=UserId(92077080),message="w")]
    KEY [UserId(78530050)@1531502760000/1531502770000] Message [Message(userId=UserId(78530050),message="t")]
    KEY [UserId(64640536)@1531502760000/1531502770000] Message [Message(userId=UserId(64640536),message="y")]

Report 3
    KEY [UserId(79117307)@1531502780000/1531502790000] Message [Message(userId=UserId(79117307),message="e")]