如何计算Java响应式Spring Boot方法的总执行耗时

问题原因 #

你之前的实现中，doFinally被附加在了flatMapMany内部的子Flux上，这会导致日志在内部子Flux完成时就触发，而不是整个verifyStuff方法返回的完整Flux执行完毕时输出。

正确实现方案 #

把doFinally移到整个Flux链的末尾，确保它监听的是整个方法返回的流的完成信号，同时保持StopWatch的初始化和启动逻辑不变：

public Flux<Void> verifyStuff(){
    StopWatch stopWatch = new StopWatch();
    stopWatch.start();
    return client.getSomeData1(query)
            .doOnNext(result1 -> log.info("Retrieved '{}' ", result1.size()))
            .flatMapMany(result1 -> Flux.fromStream(result1.stream())
                    .concatMap(this::verifyresult1))
            .doFinally(signalType -> {
                stopWatch.stop();
                log.info("Total execution time for verifying marketplace partners is {}ms",
                        stopWatch.getTotalTimeMillis());
            });
}

关于elapsed()的用法补充 #

如果需要记录每个元素从订阅到处理完成的耗时，可以用elapsed()，但它不适合直接记录整个流的总耗时：

// 示例：记录每个verifyresult1操作的耗时
public Flux<Void> verifyStuff(){
    return client.getSomeData1(query)
            .doOnNext(result1 -> log.info("Retrieved '{}' ", result1.size()))
            .flatMapMany(result1 -> Flux.fromStream(result1.stream())
                    .concatMap(item -> verifyresult1(item)
                            .elapsed()
                            .doOnNext(tuple -> log.info("Processed item in {}ms", tuple.getT1()))
                            .map(Tuple2::getT2)));
}

更贴合响应式风格的方案（可选） #

如果不想用StopWatch，可以通过AtomicLong记录起始时间，结合doFinally计算总耗时：

public Flux<Void> verifyStuff(){
    AtomicLong startTime = new AtomicLong();
    return client.getSomeData1(query)
            .doOnSubscribe(sub -> startTime.set(System.currentTimeMillis()))
            .doOnNext(result1 -> log.info("Retrieved '{}' ", result1.size()))
            .flatMapMany(result1 -> Flux.fromStream(result1.stream())
                    .concatMap(this::verifyresult1))
            .doFinally(signalType -> {
                long totalTime = System.currentTimeMillis() - startTime.get();
                log.info("Total execution time for verifying marketplace partners is {}ms", totalTime);
            });
}

内容的提问来源于stack exchange，提问作者Kartik Kodag