Skip to content

performance-testing.md

Latest commit

 

History

History
293 lines (231 loc) · 7.29 KB

performance-testing.md

File metadata and controls

293 lines (231 loc) · 7.29 KB

性能测试指南

本文档说明如何用 k6、Prometheus 和 Grafana 测 GCFeed 的接口性能、QPS、P95 延迟、错误率和缓存效果。

前置准备

安装 k6：

brew install k6

启动项目：

cd /Users/sealos/repositories/GCFeed/apps
docker compose up -d --build

确认 API 正常：

curl http://127.0.0.1:8080/health

确认监控正常：

curl http://127.0.0.1:8080/metrics
curl http://127.0.0.1:9090/-/ready
curl http://127.0.0.1:3000/api/health

Grafana 面板：

http://127.0.0.1:3000/d/gcfeed-overview/gcfeed-overview

默认账号密码：

admin / admin

测最新视频流

目标：测公开视频 Feed 的 QPS、成功率、平均延迟和 P95 延迟。

SCENE=timeline VUS=20 DURATION=60s THINK_TIME=1 k6 run - <<'EOF'
import http from "k6/http";
import { check, sleep } from "k6";
import { Rate } from "k6/metrics";

const BASE_URL = (__ENV.BASE_URL || "http://127.0.0.1:8080").replace(/\/$/, "");
const SCENE = __ENV.SCENE || "timeline";
const LIMIT = Number(__ENV.LIMIT || 10);
const THINK_TIME = Number(__ENV.THINK_TIME || 1);
const successRate = new Rate("feed_success_rate");

export const options = {
  vus: Number(__ENV.VUS || 20),
  duration: __ENV.DURATION || "60s",
  thresholds: {
    http_req_failed: ["rate<0.01"],
    http_req_duration: ["p(95)<500"],
    feed_success_rate: ["rate>0.99"],
  },
};

export default function () {
  const url = `${BASE_URL}/api/feed-items?scene=${encodeURIComponent(SCENE)}&limit=${LIMIT}`;
  const res = http.get(url);
  const ok = check(res, {
    "status is 200": (r) => r.status === 200,
    "has feed items array": (r) => Array.isArray(r.json().items),
  });
  successRate.add(ok);
  sleep(THINK_TIME);
}
EOF

重点看：

  • http_reqs 后面的 /s：QPS
  • http_req_duration avg：平均延迟
  • http_req_duration p(95)：P95 延迟
  • http_req_failed：失败率
  • feed_success_rate：业务成功率

测热门榜单

目标：测热榜读取链路，包括 Redis 热榜窗口和 Feed 卡片组装。

SCENE=hot VUS=20 DURATION=60s THINK_TIME=1 k6 run - <<'EOF'
import http from "k6/http";
import { check, sleep } from "k6";
import { Rate } from "k6/metrics";

const BASE_URL = (__ENV.BASE_URL || "http://127.0.0.1:8080").replace(/\/$/, "");
const SCENE = __ENV.SCENE || "hot";
const LIMIT = Number(__ENV.LIMIT || 10);
const THINK_TIME = Number(__ENV.THINK_TIME || 1);
const successRate = new Rate("feed_success_rate");

export const options = {
  vus: Number(__ENV.VUS || 20),
  duration: __ENV.DURATION || "60s",
  thresholds: {
    http_req_failed: ["rate<0.01"],
    http_req_duration: ["p(95)<500"],
    feed_success_rate: ["rate>0.99"],
  },
};

export default function () {
  const url = `${BASE_URL}/api/feed-items?scene=${encodeURIComponent(SCENE)}&limit=${LIMIT}`;
  const res = http.get(url);
  const ok = check(res, {
    "status is 200": (r) => r.status === 200,
    "has feed items array": (r) => Array.isArray(r.json().items),
  });
  successRate.add(ok);
  sleep(THINK_TIME);
}
EOF

测推荐流

推荐流需要登录态。先准备一个账号密码，再运行：

ACCOUNT="你的账号" PASSWORD="你的密码" VUS=20 DURATION=60s THINK_TIME=1 k6 run - <<'EOF'
import http from "k6/http";
import { check, sleep } from "k6";
import { Rate } from "k6/metrics";

const BASE_URL = (__ENV.BASE_URL || "http://127.0.0.1:8080").replace(/\/$/, "");
const ACCOUNT = __ENV.ACCOUNT || "";
const PASSWORD = __ENV.PASSWORD || "";
const LIMIT = Number(__ENV.LIMIT || 10);
const THINK_TIME = Number(__ENV.THINK_TIME || 1);
const successRate = new Rate("feed_success_rate");

export const options = {
  vus: Number(__ENV.VUS || 20),
  duration: __ENV.DURATION || "60s",
  thresholds: {
    http_req_failed: ["rate<0.01"],
    http_req_duration: ["p(95)<500"],
    feed_success_rate: ["rate>0.99"],
  },
};

export function setup() {
  const res = http.post(
    `${BASE_URL}/api/sessions`,
    JSON.stringify({ account: ACCOUNT, password: PASSWORD }),
    { headers: { "Content-Type": "application/json" } }
  );
  if (res.status !== 200) {
    throw new Error(`login failed: status=${res.status} body=${res.body}`);
  }
  return { token: res.json().access_token };
}

export default function (data) {
  const res = http.post(
    `${BASE_URL}/api/feed-queries`,
    JSON.stringify({
      scene: "recommend",
      limit: LIMIT,
      context: { request_id: `k6-recommend-${Date.now()}-${__VU}-${__ITER}` },
    }),
    {
      headers: {
        "Content-Type": "application/json",
        Authorization: `Bearer ${data.token}`,
      },
    }
  );
  const ok = check(res, {
    "status is 200": (r) => r.status === 200,
    "has feed items array": (r) => Array.isArray(r.json().items),
  });
  successRate.add(ok);
  sleep(THINK_TIME);
}
EOF

测极限 QPS

普通压测会模拟用户停顿，THINK_TIME=1 时 20 VU 的理论 QPS 接近 20。测服务极限吞吐时，把等待时间设成 0：

SCENE=timeline VUS=50 DURATION=60s THINK_TIME=0 k6 run - <<'EOF'
import http from "k6/http";
import { check, sleep } from "k6";

const BASE_URL = (__ENV.BASE_URL || "http://127.0.0.1:8080").replace(/\/$/, "");
const SCENE = __ENV.SCENE || "timeline";
const THINK_TIME = Number(__ENV.THINK_TIME || 0);

export const options = {
  vus: Number(__ENV.VUS || 50),
  duration: __ENV.DURATION || "60s",
};

export default function () {
  const res = http.get(`${BASE_URL}/api/feed-items?scene=${SCENE}&limit=10`);
  check(res, { "status is 200": (r) => r.status === 200 });
  sleep(THINK_TIME);
}
EOF

逐步增加 VUS：

VUS=50
VUS=100
VUS=200

当 P95 明显升高、失败率上升或 CPU/数据库压力明显升高时，就接近当前本地环境上限。

如何解读结果

示例：

http_reqs......................: 1200    19.85/s
http_req_duration..............: avg=5.35ms p(95)=17.96ms
http_req_failed................: 0.00%
feed_success_rate..............: 100.00%

含义：

  • http_reqs 19.85/s：QPS 约 19.85
  • avg=5.35ms：平均响应时间 5.35ms
  • p(95)=17.96ms：95% 请求在 17.96ms 内完成
  • http_req_failed=0.00%：HTTP 失败率为 0
  • feed_success_rate=100%：业务检查全部通过

可写进简历：

使用 k6 对 Feed 接口进行 20 VU / 60s 压测，完成 1200 次请求，吞吐量约 19.85 QPS，成功率 100%，错误率 0%，P95 延迟 17.96ms。

结合 Grafana 看指标

压测时打开：

http://127.0.0.1:3000/d/gcfeed-overview/gcfeed-overview

重点观察：

  • API QPS
  • API 5xx Error Rate
  • API P95 Latency
  • Feed P95 Latency
  • Feed Cache Hit Rate
  • Upload and Video Processing P95
  • Worker Success Rate

Prometheus 也可以直接查询：

sum(rate(gcfeed_http_requests_total[5m])) by (route)
histogram_quantile(0.95, sum(rate(gcfeed_http_request_duration_seconds_bucket[5m])) by (le, route))
histogram_quantile(0.95, sum(rate(gcfeed_feed_request_duration_seconds_bucket[5m])) by (le, scene))
sum(rate(gcfeed_feed_cache_requests_total{result="hit"}[5m])) by (area)

测试前的数据准备

为了让结果更接近真实场景，建议先准备：

  • 20 到 50 个公开视频
  • 多个用户
  • 一些点赞、收藏、评论
  • 至少一次推荐流访问，让推荐候选链路产生数据

当 Feed 返回空数组时，延迟指标仍然有效，业务场景说服力会弱一些。