task-99: 全链路压测记录 CPU、内存、GC、DB、Redis、RustFS、网络结果

2026-08-30 23:17:02 +08:00
parent e79f1015cb
commit 2fb56632c9
2 changed files with 355 additions and 0 deletions
@@ -0,0 +1,180 @@
+/**
+ * 全链路压测记录器（Task 99）。
+ *
+ * 执行压测请求并记录 CPU、内存、GC、DB、Redis、RustFS、网络结果：
+ * metric(i) 产生单次请求结果（ok/延迟），sample() 周期采集系统指标，
+ * 输出请求统计 + 各指标的平均/最大/min/95 分位汇总。
+ *
+ * 语义：
+ * - 并发有界：concurrency 个并发槽执行 requests 个请求，不无界扩张；
+ * - 失败有界：maxFailures 设定失败容忍上限，超过后 ok=false，但仍完成
+ *   全部请求并输出完整报告；
+ * - 采样失败（依赖故障）不中断压测：计入 sampleFailures，统计从有效
+ *   采样计算；依赖恢复后同一 recorder 重跑即全量恢复；
+ * - 空请求数安全返回；单请求不依赖批量路径；同一输入重复执行幂等，
+ *   无残留状态；
+ * - 校验失败 fail-fast：metric/sample 非函数、requests 非非负整数、
+ *   concurrency/maxFailures 非法、latencyMs 非法均抛错。
+ */
+export interface LoadMetric {
+  ok: boolean
+  output: string
+  latencyMs: number
+}
+
+export interface LoadSample {
+  cpuPercent: number
+  heapBytes: number
+  gcCount: number
+  dbQps: number
+  redisQps: number
+  rustfsQps: number
+  networkBytesPerSec: number
+}
+
+export interface LoadTestRecorderOptions {
+  /** 单次请求执行器；抛错按失败计入 */
+  metric: (i: number) => Promise<LoadMetric>
+  /** 系统指标采样器；抛错计入 sampleFailures 不中断压测 */
+  sample: () => Promise<LoadSample>
+  /** 请求总数，必须为非负整数，默认 100 */
+  requests?: number
+  /** 并发槽数量，必须为正数，默认 1 */
+  concurrency?: number
+  /** 失败容忍上限，必须为非负整数，默认 0 */
+  maxFailures?: number
+  /** 采样间隔（每 N 个请求采样一次），必须为正数，默认 1 */
+  sampleEvery?: number
+}
+
+export interface MetricSummary {
+  avg: number
+  max: number
+  min: number
+  p95: number
+}
+
+export interface LoadTestSummary {
+  ok: boolean
+  requests: number
+  failures: number
+  samples: Record<string, MetricSummary> & { avgCount: number }
+}
+
+export interface LoadTestResult {
+  ok: boolean
+  summary: LoadTestSummary
+  metrics: Array<LoadMetric & { index: number }>
+  sampleFailures: number
+}
+
+export interface LoadTestRecorder {
+  run: () => Promise<LoadTestResult>
+}
+
+const SAMPLE_KEYS: Array<keyof LoadSample> = [
+  'cpuPercent',
+  'heapBytes',
+  'gcCount',
+  'dbQps',
+  'redisQps',
+  'rustfsQps',
+  'networkBytesPerSec',
+]
+
+function p95(values: number[]): number {
+  if (values.length === 0) return 0
+  const sorted = [...values].sort((a, b) => a - b)
+  const index = Math.min(sorted.length - 1, Math.ceil(sorted.length * 0.95) - 1)
+  return sorted[index]
+}
+
+function summarize(values: number[]): MetricSummary {
+  return {
+    avg: values.length ? values.reduce((sum, v) => sum + v, 0) / values.length : 0,
+    max: values.length ? Math.max(...values) : 0,
+    min: values.length ? Math.min(...values) : 0,
+    p95: p95(values),
+  }
+}
+
+export function createLoadTestRecorder(options: LoadTestRecorderOptions): LoadTestRecorder {
+  if (typeof options.metric !== 'function') {
+    throw new Error('metric 必须是函数')
+  }
+  if (typeof options.sample !== 'function') {
+    throw new Error('sample 必须是函数')
+  }
+  const requests = options.requests ?? 100
+  const concurrency = options.concurrency ?? 1
+  const maxFailures = options.maxFailures ?? 0
+  const sampleEvery = options.sampleEvery ?? 1
+  if (!Number.isInteger(requests) || requests < 0) {
+    throw new Error('requests 必须是非负整数: ' + requests)
+  }
+  if (!(concurrency > 0)) {
+    throw new Error('concurrency 必须为正数: ' + concurrency)
+  }
+  if (!Number.isInteger(maxFailures) || maxFailures < 0) {
+    throw new Error('maxFailures 必须是非负整数: ' + maxFailures)
+  }
+  if (!(sampleEvery > 0)) {
+    throw new Error('sampleEvery 必须为正数: ' + sampleEvery)
+  }
+
+  async function run(): Promise<LoadTestResult> {
+    const metrics: Array<LoadMetric & { index: number }> = []
+    const samplesByKey: Record<string, number[]> = {}
+    for (const key of SAMPLE_KEYS) samplesByKey[key] = []
+    let sampleFailures = 0
+
+    const runRequest = async (i: number) => {
+      let metric: LoadMetric
+      try {
+        metric = await options.metric(i)
+      } catch (error) {
+        metric = { ok: false, output: error instanceof Error ? error.message : String(error), latencyMs: 0 }
+      }
+      if (typeof metric.latencyMs !== 'number' || !Number.isFinite(metric.latencyMs) || metric.latencyMs < 0) {
+        throw new Error('latencyMs 必须为非负数值: ' + metric.latencyMs)
+      }
+      metrics.push({ ...metric, index: i })
+      if (i % sampleEvery === 0) {
+        try {
+          const sample = await options.sample()
+          for (const key of SAMPLE_KEYS) {
+            samplesByKey[key].push(sample[key])
+          }
+        } catch {
+          sampleFailures += 1
+        }
+      }
+    }
+
+    for (let offset = 0; offset < requests; offset += concurrency) {
+      const batch = []
+      for (let i = offset; i < Math.min(offset + concurrency, requests); i++) {
+        batch.push(runRequest(i))
+      }
+      await Promise.all(batch)
+    }
+
+    metrics.sort((a, b) => a.index - b.index)
+    const failures = metrics.filter((m) => !m.ok).length
+    const summary: LoadTestSummary = {
+      ok: failures <= maxFailures,
+      requests: metrics.length,
+      failures,
+      samples: {
+        avgCount: samplesByKey.cpuPercent.length,
+      } as LoadTestSummary['samples'],
+    }
+    for (const key of SAMPLE_KEYS) {
+      ;(summary.samples as Record<string, MetricSummary>)[key] = summarize(samplesByKey[key])
+    }
+
+    return { ok: failures <= maxFailures, summary, metrics, sampleFailures }
+  }
+
+  return { run }
+}