degist-concurrency-bugfix

[wxbty]

What is the purpose of the change

Issue Number: close #12153

Brief changelog

The MergingDigest data structure of the t-Digest package uses a cache with a default size (1050) to improve efficiency, and the cache increments by +1 each time it is added.
If the cache is full, the space will be merged and tempUsed=0 will be reset. In this process, there are common concurrency problems in the judgment and merge methods, which need to be fixed.

1. Add synchronized to the mergeNewValues ​​method. This method will be executed about 1050 times (the reason for the 0.1% error rate), so there is no need to worry about performance issues.
2. Add volatile modification
3. Because the source code of t-Digest cannot be modified directly, copy the corresponding code to the dubbo source code separately

 private void add(double x, int w, List<Double> history) {
        if (Double.isNaN(x)) {
            throw new IllegalArgumentException("Cannot add NaN to t-digest");
        }
        if (tempUsed >= tempWeight.length - lastUsedCell - 1) {  //判断是否达到缓存上限，tempWeight.length =1050
            mergeNewValues();
        }
        int where = tempUsed++;    // 每次add自增
        // ...
}

Verifying this change

see testcase testMulti, also as follow:
1、git pull 3.2
2、modify the ConsumerApplication code of dubbo-demo-spring-boot-consumer as follows. About 0.1% error rate before the change, no error rate after the change

 public static void main(String[] args) {

        ConfigurableApplicationContext context = SpringApplication.run(ConsumerApplication.class, args);
        ConsumerApplication application = context.getBean(ConsumerApplication.class);
        String result = application.doSayHello();
        System.out.println("result: " + result);
    }

    public String doSayHello() {
        int index = 0;
        while (index < 200) {
            System.out.println("-------------:" + index);
            for (int i = 0; i < 100; i++) {
                new Thread(() -> {

                    try {
                        String result = demoService.sayHello("world");
                    } catch (Exception e) {
                        e.printStackTrace();
                    }
                }).start();

            }
            index++;
            try {
                Thread.sleep(500);
            } catch (InterruptedException e) {
                e.printStackTrace();
            }
        }
        return "";
    }

[codecov-commenter]

Codecov Report

Merging #12223 (8c128c0) into 3.2 (c5f8ce6) will increase coverage by 0.58%.
The diff coverage is n/a.

@@             Coverage Diff              @@
##                3.2   #12223      +/-   ##
============================================
+ Coverage     68.82%   69.40%   +0.58%     
+ Complexity      115        2     -113     
============================================
  Files          3559     1605    -1954     
  Lines        163607    66165   -97442     
  Branches      26766     9698   -17068     
============================================
- Hits         112597    45922   -66675     
+ Misses        41209    15815   -25394     
+ Partials       9801     4428    -5373     

see 1986 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[AlbumenJ]

Is it possible to add some uts to verify it?

[AlbumenJ]

@songxiaosheng PTAL

[songxiaosheng]

是否可以考虑比micrometer的滑动窗口函数，来实现相同的功能
Micrometer是一个度量库，它提供了许多度量指标，包括计数器、计时器、直方图等等，可以用于监控应用程序的性能和健康状况。而滑动窗口是一种常用的度量指标的实现方式，它可以用来统计一段时间内的度量指标数据，比如最近一分钟、最近一小时等等。

Micrometer提供了一个名为SlidingWindow的滑动窗口实现，可以用来统计一段时间内的度量指标数据。该实现可以通过以下方式创建：

java
Copy code
SlidingWindow slidingWindow = new SlidingWindow(TimeUnit.SECONDS, 10);

这里创建了一个10秒的滑动窗口。可以使用record方法向滑动窗口中添加度量指标数据，例如：

java
Copy code
Counter counter = Counter.builder("my.counter").register(registry);
slidingWindow.record(counter::count);

这里创建了一个名为my.counter的计数器，并将其记录到滑动窗口中。可以使用stream方法获取滑动窗口中的度量指标数据，例如：

java
Copy code
List measurements = slidingWindow.stream().map(Measurement::getValue).collect(Collectors.toList());

这里获取了滑动窗口中所有度量指标数据的值，并将其存储在一个List中。可以根据需要对这些数据进行进一步处理，例如计算平均值、最大值、最小值等等。

[AlbumenJ]

是否可以考虑比micrometer的滑动窗口函数，来实现相同的功能 Micrometer是一个度量库，它提供了许多度量指标，包括计数器、计时器、直方图等等，可以用于监控应用程序的性能和健康状况。而滑动窗口是一种常用的度量指标的实现方式，它可以用来统计一段时间内的度量指标数据，比如最近一分钟、最近一小时等等。

Micrometer提供了一个名为SlidingWindow的滑动窗口实现，可以用来统计一段时间内的度量指标数据。该实现可以通过以下方式创建：

java Copy code SlidingWindow slidingWindow = new SlidingWindow(TimeUnit.SECONDS, 10);

这里创建了一个10秒的滑动窗口。可以使用record方法向滑动窗口中添加度量指标数据，例如：

java Copy code Counter counter = Counter.builder("my.counter").register(registry); slidingWindow.record(counter::count);

这里创建了一个名为my.counter的计数器，并将其记录到滑动窗口中。可以使用stream方法获取滑动窗口中的度量指标数据，例如：

java Copy code List measurements = slidingWindow.stream().map(Measurement::getValue).collect(Collectors.toList());

这里获取了滑动窗口中所有度量指标数据的值，并将其存储在一个List中。可以根据需要对这些数据进行进一步处理，例如计算平均值、最大值、最小值等等。

支持，我们可以在这个 PR 修复完以后尝试基于 micrometer 进行重构

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
23 Code Smells

43.1% Coverage
0.0% Duplication

[songxiaosheng]

LGTM