以为是大厂,再怎么也不会差到哪里去。结果用它加速代理后的服务器有 80% 是不可达的状态。。。
#### 2. 原始 TCP 443 探测:13/16 不可达
361 +
362 +```
363 +163.181.78.144 FAIL (Operation timed out)
364 +163.181.78.145 FAIL (Operation timed out)
365 +163.181.78.146 FAIL (Operation timed out)
366 +163.181.78.147 OK
367 +163.181.78.148 OK
368 +163.181.78.161 FAIL (Operation timed out)
369 +163.181.78.162 FAIL (Operation timed out)
370 +163.181.78.163 FAIL (Operation timed out)
371 +163.181.78.164 FAIL (Operation timed out)
372 +163.181.78.165 FAIL (Operation timed out)
373 +163.181.78.167 FAIL (Operation timed out)
374 +163.181.78.168 OK ← 2026-05-13 还是 FAIL,今天活了
375 +163.181.78.169 FAIL (Operation timed out) ← 2026-05-12 是 OK,2026-05-13 起死
376 +163.181.78.170 FAIL (Operation timed out)
377 +163.181.78.171 FAIL (Operation timed out)
378 +163.181.78.172 FAIL (Operation timed out)
379 +
380 +OK=3 FAIL=13 total=16 (81.25% 不可达)
381 +```
382 +
383 +可用 IP 集合从 `{.147 .148 .169}` 漂移到 `{.147 .148 .168}`,**池内单点偶有抖动,但整体死亡率稳定在 80%+ 量级**。
是阿里云 ESA(免费版)侧的问题,不是代码、也不是 ESA 控制台规则问题。 阿里云 ESA 给 xxx.com 分配的 16 个边缘 IP 里,13 个在 raw TCP 层完全不可达 (连 SYN-ACK 都不回),浏览器随机命中导致冷连接 TCP 阶段被卡 0.6–1 秒甚至超时。
2 个帖子 - 2 位参与者