Implementing Response Time Statistics in the FunTester Framework

It is known that, aside from JVM warm‑up issues, some requests exhibit abnormal response times; the article shares a case where filtering the initial response time records improves overall data accuracy.

The need for optimization arose because the FunTester testing framework originally did not collect local statistics, yet after a recent JDK upgrade and GC parameter comparisons, monitoring data from gateways differed from real user scenarios, prompting the addition of client‑side response‑time tracking.

Response times are stored as a short array: for single‑threaded cases in a List<Short> costs and for load‑test cases in a Vector<Short> . A generic method static FunIndex index(List c) sorts the list and extracts various quantiles (p99, p999, p95), minimum, maximum, average, and median values. The implementation is shown below:

/**
 * 统计list各分位数据
 * @param c
 * @return
 */
static FunIndex index(List
c) {
    if (c == null || c.size() == 0) return
    c.sort()
    int size = c.size()
    double min = c.first()
    double max = c.last()
    double p99 = c.get(size * 0.99 as Integer)
    double p999 = c.get(size * 0.999 as Integer)
    double p95 = c.get(size * 0.95 as Integer)
    double avg = SourceCode.changeStringToDouble(SourceCode.formatNumber(c.average(), "#.###"))
    def mid = c.get(size / 2 as Integer)
    new FunIndex(avg: avg, mid: mid, min: min, max: max, p99: p99, p999: p999, p95: p95)
}

/**
 * 统计结果
 */
static class FunIndex extends AbstractBean {
    Double avg
    Double mid
    Double min
    Double max
    Double p99
    Double p999
    Double p95
    @Override
    String toString() {
        "平均值:$avg ,最大值$max ,最小值:$min ,中位数:$mid p99:$p99 p95:$p95"
    }
}

For each thread, the per‑thread List<Short> costs is removed and the data is centralized in com.funtester.base.constaint.ThreadBase . The following code shows the new flag and counting method:

/**
 * 是否记录响应时间,默认否
 */
public static boolean COUNT = false;

/**
 * 记录响应时间
 *
 * @param s 开始时间
 */
public void count(long s) {
    if (COUNT && executeNum > 100) costs.add((short) (Time.getTimeStamp() - s));
}

The thread model’s run method is also refactored to record timestamps, log execution statistics, and aggregate costs into Concurrent.allTimes . The revised method is:

@Override
public void run() {
    try {
        before();
        long ss = Time.getTimeStamp();
        while (true) {
            try {
                executeNum++;
                long s = Time.getTimeStamp();
                doing();
                count(s);
            } catch (Exception e) {
                logger.warn("执行任务失败！", e);
                errorNum++;
            } finally {
                if ((isTimesMode ? executeNum >= limit : (Time.getTimeStamp() - ss) >= limit) || ThreadBase.needAbort() || status())
                    break;
            }
            long ee = Time.getTimeStamp();
            if ((ee - ss) / 1000 > RUNUP_TIME + 3) // 区分软启动运行和正式运行
                logger.info("线程:{},执行次数：{},错误次数: {},总耗时：{} s", threadName, executeNum, errorNum, (ee - ss) / 1000.0);
            Concurrent.allTimes.addAll(costs);
            Concurrent.requestMark.addAll(marks);
        }
    } catch (Exception e) {
        logger.warn("执行任务失败！", e);
    } finally {
        after();
    }
}

Following this approach, the author plans to add an asynchronous RT sampler based on the same idea, linking to a dynamic model article for future reference.

Have Fun ~ Tester ！