fix gateway probe bad port transport

2026-05-25 03:29:52 +00:00
parent 57eb380247
commit 77e73f8bda
5 changed files with 93 additions and 32 deletions
@@ -57,6 +57,8 @@ NODE
 - 不要把多层脚本内联穿过 `ssh -> bash -lc -> kubectl exec -> sh -lc -> node/python/powershell`。这种路径会让 heredoc、`$!`、反引号、JSON 字符串、PowerShell 管道和中文内容被不同 shell 多次解释，常见结果是实验命令还没进入 pod 就被截断或改写。
 - 可靠做法是把实验逻辑沉淀为 repo-owned 脚本，或先写入远端 `/tmp/*.mjs|*.sh|*.ps1`，再用 `kubectl cp`/ConfigMap 挂载到 pod，最后用 `kubectl exec -- node /tmp/script.mjs --arg value` 或等价 argv 形态执行。只允许在外层 shell 保留短命令和固定参数，不在命令行里嵌大段 JS/PowerShell/heredoc。
 - 需要从 pod 内验证同一份实验时，优先复用同一个脚本在 D601、pod 和 CI 中运行；不要把探测逻辑在每个通道手写一遍。若必须临时写脚本，文件名包含目标和 commit，issue 中记录脚本路径、commit、命令和输出摘要。
+- pod 内部、D601 host、FRP 公网入口可能使用不同端口语义。`127.0.0.1:6667` 是 HWLAB 内部 cloud-api 常用端口，但 `6667` 属于 WHATWG bad port；Node/undici `fetch` 会在发包前直接报 `bad port`，包括临时 gateway 自己 poll `http://127.0.0.1:6667` 的场景。运行面实验脚本和 gateway 传输层必须用 Node `http/https` 原生 request，或改走公网 `http://74.48.78.17:16667` / service mesh 中不被 bad-port 拦截的入口。
+- 遇到 `fetch failed`、`bad port`、公网通而 pod loopback 不通时，先判断是不是实验工具栈限制，不要立即归因 cloud-api、FRP、k3s Service 或 gateway 离线。issue 复盘要记录具体 URL、调用库、错误字符串和替代入口。
 - 不要对已经很大的 hotfix ConfigMap 使用 `kubectl apply -f -`；`apply` 会尝试写入 `kubectl.kubernetes.io/last-applied-configuration`，可能超过 Kubernetes annotation 256 KiB 限制。局部更新用 `patch --type merge --patch-file`。
 - 多文件运行面覆盖第一次创建 ConfigMap 时也不要默认用 `kubectl apply -f -`；如果 data 可能超过 annotation 限制，先 `kubectl delete configmap <name> --ignore-not-found`，再 `kubectl create configmap <name> --from-file=...`，随后用 Deployment annotation 记录 hotfix 版本并 rollout。正式 CD 收口时必须删除这些 unmanaged ConfigMap/volumeMount，不让热修覆盖继续漂在运行面。
 - 不要通过会截断 stdout 的控制 CLI 把 `kubectl get configmap -o json` 回读到本地再 `replace`；大 data 可能被日志/截断污染成非法 JSON。只有在确认完整无截断的原生通道中，才允许对完整对象做 `kubectl replace -f -`。
@@ -60,7 +60,7 @@ npm run gateway:demo:edge-smoke
 - `gateway:demo:smoke` 启动本地 `hwlab-cloud-api` 和 `hwlab-gateway`，验证 `hardware.invoke.shell` 返回 `stdout=hwlab-demo`。
 - `gateway:demo:edge-smoke` 额外启动本地 `hwlab-edge-proxy`，验证普通 HTTP proxy 能转发 `/v1/gateway/poll`、`/v1/gateway/result` 和 `/json-rpc`。
 - 两个 smoke 都会设置 `NO_PROXY/no_proxy`，用于规避本地代理误触发。
- 对 D601 pod 内部 `127.0.0.1:6667` 或 service `:6667` 做 Node 探测时，不要用 Web/undici `fetch`；`6667` 属于 WHATWG bad port，Node 会直接报 `bad port`。repo-owned 探测脚本应使用 `http/https` 原生 request，公网 `:16667` 或浏览器入口不受这个限制。
+- 对 D601 pod 内部 `127.0.0.1:6667` 或 service `:6667` 做 Node 探测时，不要用 Web/undici `fetch`；`6667` 属于 WHATWG bad port，Node 会直接报 `bad port`。gateway 传输层和 repo-owned 探测脚本应使用 `http/https` 原生 request，公网 `:16667` 或浏览器入口不受这个限制。

 ## DEV 使用

@@ -1,4 +1,5 @@
-import { createServer } from "node:http";
+import { createServer, request as httpRequest } from "node:http";
+import { request as httpsRequest } from "node:https";

 export class HttpError extends Error {
  constructor(statusCode, body) {
@@ -76,28 +77,47 @@ export function createJsonServer({ routes }) {

 export async function fetchJson(url, options = {}) {
  const body = options.body === undefined ? undefined : JSON.stringify(options.body);
-  const response = await fetch(url, {
-    method: options.method ?? (body === undefined ? "GET" : "POST"),
-    headers: {
-      ...(body === undefined
-        ? {}
-        : {
-            "content-type": "application/json"
-          }),
-      ...(options.headers ?? {})
-    },
-    body,
-    signal: options.signal
-  });
-  const text = await response.text();
-  const json = text ? JSON.parse(text) : null;
-
-  return {
-    ok: response.ok,
-    status: response.status,
-    statusText: response.statusText,
-    body: json
+  const target = url instanceof URL ? url : new URL(String(url));
+  const client = target.protocol === "https:" ? httpsRequest : httpRequest;
+  const headers = {
+    ...(body === undefined
+      ? {}
+      : {
+          "content-type": "application/json",
+          "content-length": Buffer.byteLength(body)
+        }),
+    ...(options.headers ?? {})
  };
+  return new Promise((resolve, reject) => {
+    const request = client(target, {
+      method: options.method ?? (body === undefined ? "GET" : "POST"),
+      headers,
+      signal: options.signal
+    }, (response) => {
+      let text = "";
+      response.setEncoding("utf8");
+      response.on("data", (chunk) => {
+        text += chunk;
+      });
+      response.on("end", () => {
+        try {
+          const json = text ? JSON.parse(text) : null;
+          const status = response.statusCode ?? 0;
+          resolve({
+            ok: status >= 200 && status < 300,
+            status,
+            statusText: response.statusMessage ?? "",
+            body: json
+          });
+        } catch (error) {
+          reject(error);
+        }
+      });
+    });
+    request.on("error", reject);
+    if (body !== undefined) request.write(body);
+    request.end();
+  });
 }

 export function parsePort(value, fallback) {
@@ -1,6 +1,8 @@
 #!/usr/bin/env node
 import { spawn } from "node:child_process";
 import { createServer } from "node:net";
+import http from "node:http";
+import https from "node:https";
 import { setTimeout as delay } from "node:timers/promises";
 import assert from "node:assert/strict";

@@ -183,26 +185,56 @@ async function waitForJson(url) {
 }

 async function getJson(url) {
-  const response = await fetch(url, {
+  return requestJson(url, {
+    method: "GET",
    headers: { accept: "application/json" }
  });
-  const text = await response.text();
-  if (!response.ok) throw new Error(`HTTP ${response.status}: ${text}`);
-  return text ? JSON.parse(text) : null;
 }

 async function postJson(url, body) {
-  const response = await fetch(url, {
+  return requestJson(url, {
    method: "POST",
    headers: {
      accept: "application/json",
      "content-type": "application/json"
    },
-    body: JSON.stringify(body)
+    body
+  });
+}
+
+function requestJson(url, { method, headers = {}, body }) {
+  return new Promise((resolve, reject) => {
+    const target = new URL(url);
+    const client = target.protocol === "https:" ? https : http;
+    const payload = body === undefined ? "" : JSON.stringify(body);
+    const request = client.request(target, {
+      method,
+      headers: {
+        ...headers,
+        ...(payload ? { "content-length": Buffer.byteLength(payload) } : {})
+      }
+    }, (response) => {
+      let text = "";
+      response.setEncoding("utf8");
+      response.on("data", (chunk) => {
+        text += chunk;
+      });
+      response.on("end", () => {
+        if ((response.statusCode ?? 0) < 200 || (response.statusCode ?? 0) >= 300) {
+          reject(new Error(`HTTP ${response.statusCode}: ${text}`));
+          return;
+        }
+        try {
+          resolve(text ? JSON.parse(text) : null);
+        } catch (error) {
+          reject(new Error(`non-JSON response: ${error.message}`));
+        }
+      });
+    });
+    request.on("error", reject);
+    if (payload) request.write(payload);
+    request.end();
  });
-  const text = await response.text();
-  if (!response.ok) throw new Error(`HTTP ${response.status}: ${text}`);
-  return text ? JSON.parse(text) : null;
 }

 function shellRpc({ id, traceId, command, cwd, timeoutMs }) {
@@ -72,6 +72,7 @@ const gatewayDemoRegistry = fs.readFileSync(path.resolve(repoRoot, "internal/clo
 const gatewayMain = fs.readFileSync(path.resolve(repoRoot, "cmd/hwlab-gateway/main.mjs"), "utf8");
 const gatewayDemoSmoke = fs.readFileSync(path.resolve(repoRoot, "scripts/gateway-outbound-demo-smoke.mjs"), "utf8");
 const gatewayNonblockingProbe = fs.readFileSync(path.resolve(repoRoot, "scripts/gateway-outbound-nonblocking-probe.mjs"), "utf8");
+const simHttp = fs.readFileSync(path.resolve(repoRoot, "internal/sim/http.mjs"), "utf8");
 const codexStdioSession = fs.readFileSync(path.resolve(repoRoot, "internal/cloud/codex-stdio-session.mjs"), "utf8");
 const gatewayShellTool = fs.readFileSync(path.resolve(repoRoot, "tools/hwlab-gateway-shell.mjs"), "utf8");
 const frontendSource = `${html}\n${auth}\n${app}\n${artifactPublisher}`;
@@ -1163,7 +1164,13 @@ assert.match(gatewayMain, /function startCloudRequest/);
 assert.match(gatewayMain, /Promise\.resolve\(\)\s*\n\s*\.then\(\(\) => handleCloudRequest/);
 assert.doesNotMatch(gatewayMain, /await handleCloudRequest\(response\.body\.request\)/);
 assert.match(gatewayDemoSmoke, /quick request was blocked by slow request/);
+assert.match(gatewayDemoSmoke, /function requestJson/);
+assert.doesNotMatch(functionBody(gatewayDemoSmoke, "getJson"), /\bfetch\(/);
+assert.doesNotMatch(functionBody(gatewayDemoSmoke, "postJson"), /\bfetch\(/);
 assert.match(gatewayNonblockingProbe, /quick request was blocked by slow request/);
+assert.match(simHttp, /request as httpRequest/);
+assert.match(functionBody(simHttp, "fetchJson"), /httpRequest/);
+assert.doesNotMatch(functionBody(simHttp, "fetchJson"), /\bfetch\(/);
 assert.match(cloudApiServer, /HWLAB_GATEWAY_DISPATCH_TIMEOUT_MS,\s*120000/);
 assert.match(codexStdioSession, /detached:\s*childDetached/);
 assert.match(codexStdioSession, /function terminateChildProcess/);