Upload 21 files

.env.example

@@ -10,7 +10,7 @@ CODEX_MODEL=your-model
 # Optional
 CODEX_REASONING_EFFORT=high
 # Space container forces danger-full-access inside Docker to avoid bwrap namespaces.
-CODEX_TIMEOUT_MS=300000
+CODEX_TIMEOUT_MS=900000
 MAX_REPLY_CHARS=3500
 HISTORY_TURNS=8
 MAX_INPUT_FILE_BYTES=20971520

DEPLOY.md

@@ -35,3 +35,15 @@ Space 上线并取得真实公开域名后，在 UptimeRobot 选择 HTTP(s) moni
 Hugging Face Space 的 Docker 运行环境可能禁止非特权 user namespace。Codex 的 `workspace-write` 沙箱依赖 bwrap，会报 `No permissions to create a new namespace`，导致所有工具命令执行失败，即使模型 API 和 curl 本身可访问。QQ 桥接现固定调用 `codex exec -s danger-full-access`，不再启动 bwrap；Jupyter Terminal 的 Codex 配置也默认 `sandbox_mode = "danger-full-access"`。原先 Space Settings 中的 `CODEX_SANDBOX=workspace-write` 可删除，桥接不会再读取它。**需要重新上传并重新构建 Space**；旧镜像仍会使用旧代码。
 
 这仅依赖 Docker/Space 容器的隔离，不等同于 Codex 的文件系统沙箱。QQ 白名单用户可以让模型读取容器内任何可读文件、执行命令、访问运行时环境变量和外网。严格保护 QQ_ALLOWED_OPENIDS 与 JUPYTER_TOKEN；不要把其他生产密钥放入同一 Space。如不能接受该风险，请改用支持非特权 user namespace 的宿主环境，或等待平台支持相应沙箱能力。
+
+
+## 长任务无回复与 Jupyter 拒绝访问排查（本版）
+旧版 `codex exec --json` 的 stdout 没有消费，输出事件积累后管道会满，使子进程卡住。新版持续消费 stdout、记录退出状态；默认 CODEX_TIMEOUT_MS=900000（15 分钟），超时后强制终止进程。若 Space Settings 已配置旧值 CODEX_TIMEOUT_MS=300000，旧值仍优先，请手动改为 900000 或 1200000（20 分钟）。QQ 每 3 分钟发送进度；发送 `/status` 可查询当前任务运行时间，不会排在任务后面。
+
+启动日志在 Space 的运行日志中输出，并同时写在容器临时目录：`/home/node/app/data/startup.log`、`jupyter.log`、`bridge.log`、`nginx.log`。启动顺序改为 Jupyter → HTTP 代理 → QQ 机器人；QQ 配置缺失或机器人退出时不会一并杀死 Jupyter 界面。`/health` 现在只确认 Jupyter 可连接，不代表 QQ 机器人已登录；通过 QQ `/status` 检查机器人。
+
+JupyterLab 必须设置 Space Secret `JUPYTER_TOKEN`，打开实际 Space App 地址的 `/lab`，在登录页面输入该 token。若显示 Hugging Face 的 403/拒绝访问页面而不是 Jupyter 登录页，检查 Space 是否 Private、账号是否有该 Space 访问权、是否打开了实际 App 地址；这类平台访问控制不是容器内 token 可以解决的。若 `/health` 为 200、`/lab` 跳转到 `/login`，说明容器与代理已工作，需检查 token/浏览器 Cookie。不要将 token 作为 URL 参数公开分享。
+
+若仍无法进入界面，请从 Hugging Face Space 的 Runtime logs 复制 `jupyter.log`/`nginx` 对应错误、以及浏览器网络请求的 HTTP 状态码（不要包含 token 或 API key）。
+
+QQ 白名单用户还可以发送 `/diag` 获取脱敏后的最近 bridge/startup/jupyter 日志；若机器人仍在线，优先用 `/status` 和 `/diag` 排查，无须进入 Jupyter Terminal。不要把诊断消息转发给其他人。容器的 Jupyter 与 QQ 日志默认保存在临时 data 目录；启动崩溃时会将脱敏后的最后几行写入 Space Runtime logs。

scripts/health.py

@@ -1,24 +1,30 @@
-import http.server
-import os
-import socket
-import sys
-PIDS = [int(p) for p in sys.argv[1:]]
-class Handler(http.server.BaseHTTPRequestHandler):
-    def do_GET(self):
-        if self.path != "/health":
-            self.send_error(404)
-            return
-        try:
-            for pid in PIDS: os.kill(pid, 0)
-            with socket.create_connection(("127.0.0.1", 8888), timeout=2): pass
-            status, body = 200, b"ok\n"
-        except OSError:
-            status, body = 503, b"unavailable\n"
-        self.send_response(status)
-        self.send_header("Content-Type", "text/plain")
-        self.send_header("Cache-Control", "no-store")
-        self.send_header("Content-Length", str(len(body)))
-        self.end_headers()
-        self.wfile.write(body)
-    def log_message(self, *_): pass
+import http.server
+import os
+import socket
+import sys
+
+PIDS = [int(p) for p in sys.argv[1:] if p.isdigit() and int(p) > 0]
+
+class Handler(http.server.BaseHTTPRequestHandler):
+    def do_GET(self):
+        if self.path != "/health":
+            self.send_error(404)
+            return
+        try:
+            for pid in PIDS:
+                os.kill(pid, 0)
+            with socket.create_connection(("127.0.0.1", 8888), timeout=2):
+                pass
+            status, body = 200, b"ok\n"
+        except OSError:
+            status, body = 503, b"unavailable\n"
+        self.send_response(status)
+        self.send_header("Content-Type", "text/plain")
+        self.send_header("Cache-Control", "no-store")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+    def log_message(self, *_):
+        pass
+
 http.server.ThreadingHTTPServer(("127.0.0.1", 7861), Handler).serve_forever()

scripts/show_logs.py

@@ -0,0 +1,16 @@
+import os
+import sys
+from pathlib import Path
+
+for filename in sys.argv[1:]:
+    path = Path(filename)
+    if not path.is_file():
+        continue
+    lines = path.read_text(encoding="utf-8", errors="replace").splitlines()[-40:]
+    secrets = [os.environ.get(name, "") for name in ("JUPYTER_TOKEN", "CODEX_API_KEY", "QQBOT_APP_SECRET", "QQBOT_APP_ID")]
+    print(f"--- {path.name} (last {len(lines)} lines, secrets redacted) ---")
+    for line in lines:
+        for secret in secrets:
+            if secret:
+                line = line.replace(secret, "[REDACTED]")
+        print(line[:1000])

scripts/start.sh

@@ -1,32 +1,60 @@
 #!/usr/bin/env bash
-set -euo pipefail
+set -Eeuo pipefail
 cd /home/node/app
-: ${QQBOT_APP_ID:?missing QQBOT_APP_ID}
-: ${QQBOT_APP_SECRET:?missing QQBOT_APP_SECRET}
-: ${QQ_ALLOWED_OPENIDS:?missing QQ_ALLOWED_OPENIDS}
-: ${CODEX_BASE_URL:?missing CODEX_BASE_URL}
-: ${CODEX_API_KEY:?missing CODEX_API_KEY}
-: ${CODEX_MODEL:?missing CODEX_MODEL}
-: ${JUPYTER_TOKEN:?missing JUPYTER_TOKEN}
 mkdir -p data workspace /tmp/nginx
-# Both chat history and Codex/Jupyter working files are ephemeral container-layer data.
+# Both chat history and Codex/Jupyter working files are ephemeral.
 find workspace -mindepth 1 -maxdepth 1 -exec rm -rf -- {} +
-rm -f data/history.json data/bridge.log data/codex-*.txt
+rm -f data/history.json data/bridge.log data/codex-*.txt data/jupyter.log data/nginx.log data/startup.log
+exec > >(tee -a data/startup.log) 2>&1
+: "${JUPYTER_TOKEN:?Set JUPYTER_TOKEN in Space Secrets to access JupyterLab}"
+printf '%s startup begin\n' "$(date -Is)"
+
 export CODEX_HOME=/home/node/app/data/codex-home
 mkdir -p "$CODEX_HOME/skills"
 cp -r /home/node/app/skills/. "$CODEX_HOME/skills/"
-python3 scripts/configure_codex.py
-codex --version
-node dist/index.js &
-bot_pid=$!
-jupyter lab --ip=127.0.0.1 --port=8888 --no-browser --ServerApp.root_dir=/home/node/app/workspace --ServerApp.token="${JUPYTER_TOKEN}" --ServerApp.allow_remote_access=True --ServerApp.trust_xheaders=True &
+
+# Start Jupyter first. A bad bot/model configuration must not make the Space UI unavailable.
+jupyter lab --ip=127.0.0.1 --port=8888 --no-browser \
+  --ServerApp.root_dir=/home/node/app/workspace \
+  --ServerApp.token="${JUPYTER_TOKEN}" \
+  --ServerApp.password='' \
+  --ServerApp.allow_remote_access=True \
+  --ServerApp.trust_xheaders=True \
+  --ServerApp.base_url=/ > data/jupyter.log 2>&1 &
 jupyter_pid=$!
-cleanup() { trap - EXIT INT TERM; kill ${bot_pid:-} ${jupyter_pid:-} ${health_pid:-} ${nginx_pid:-} 2>/dev/null || true; wait 2>/dev/null || true; }
-trap cleanup EXIT
-trap 'exit 0' INT TERM
-python3 scripts/health.py $bot_pid $jupyter_pid &
+printf '%s jupyter pid=%s\n' "$(date -Is)" "$jupyter_pid"
+
+python3 scripts/health.py "$jupyter_pid" > data/health.log 2>&1 &
 health_pid=$!
-nginx -c /home/node/app/nginx.conf -g 'daemon off;' &
+nginx -c /home/node/app/nginx.conf -g 'daemon off;' > data/nginx.log 2>&1 &
 nginx_pid=$!
-wait -n $bot_pid $jupyter_pid $nginx_pid
+
+bot_pid=''
+required=(QQBOT_APP_ID QQBOT_APP_SECRET QQ_ALLOWED_OPENIDS CODEX_BASE_URL CODEX_API_KEY CODEX_MODEL JUPYTER_TOKEN)
+missing=()
+for name in "${required[@]}"; do [[ -n "${!name:-}" ]] || missing+=("$name"); done
+if ((${#missing[@]})); then
+  printf '%s bot not started; missing variables: %s\n' "$(date -Is)" "${missing[*]}" | tee -a data/bridge.log
+else
+  python3 scripts/configure_codex.py
+  codex --version | tee -a data/bridge.log
+  node dist/index.js > data/bridge.log 2>&1 &
+  bot_pid=$!
+  printf '%s bot pid=%s started\n' "$(date -Is)" "$bot_pid" | tee -a data/bridge.log
+fi
+
+cleanup() {
+  trap - EXIT INT TERM
+  for pid in "$nginx_pid" "$health_pid" "$bot_pid" "$jupyter_pid"; do
+    [[ -n "$pid" ]] && kill "$pid" 2>/dev/null || true
+  done
+  wait 2>/dev/null || true
+}
+trap cleanup EXIT
+trap 'exit 0' INT TERM
+while kill -0 "$jupyter_pid" 2>/dev/null && kill -0 "$nginx_pid" 2>/dev/null; do
+  sleep 5
+done
+printf '%s jupyter or nginx exited; container will restart\n' "$(date -Is)"
+python3 scripts/show_logs.py data/jupyter.log data/nginx.log data/bridge.log
 exit 1

src/codex.ts

@@ -52,7 +52,44 @@ export async function askCodex(k: string, text: string, files: string[] = [], be
   const model = env("CODEX_MODEL"); if (model) args.push("-m", model); const effort = env("CODEX_REASONING_EFFORT"); if (effort) args.push("-c", `model_reasoning_effort=${JSON.stringify(effort)}`);
   for (const f of files) if (/\.(png|jpe?g|webp|gif|bmp)$/i.test(f)) args.push("-i", f); args.push("-"); await log(`request key=${k} chars=${text.length} files=${files.length}`);
   const { spawn } = await import("node:child_process"); const configured = env("CODEX_COMMAND", "codex"); const command = /^[^\\/:]+\.cmd$/i.test(configured) ? configured.replace(/\.cmd$/i, ".exe") : configured; const childEnv = await providerEnv();
-  return new Promise((resolve, reject) => { const c = spawn(command, args, { cwd: workdir, env: childEnv, windowsHide: true, stdio: ["pipe", "pipe", "pipe"] }); let err = ""; const timer = setTimeout(() => { c.kill(); reject(new Error("Codex 响应超时")); }, Number(env("CODEX_TIMEOUT_MS", "300000"))); c.stderr.on("data", b => err += String(b)); c.on("error", e => { clearTimeout(timer); reject(new Error(`无法启动 Codex：${e.message}`)); }); c.on("close", async code => { clearTimeout(timer); let answer = ""; try { answer = (await readFile(out, "utf8")).trim(); } catch {} if (code !== 0) { await log(`error code=${code} stderr=${err.slice(-1000)}`); reject(new Error(err.trim().slice(-1000) || `Codex 退出码 ${code}`)); return; } if (!answer) answer = "Codex 没有返回文本。"; const a = store[k] ??= []; a.push({ role: "user", text, at: new Date().toISOString() }, { role: "assistant", text: answer, at: new Date().toISOString() }); if (a.length > 40) a.splice(0, a.length - 40); await save(); const artifacts = await findNewArtifacts(workdir, before ?? new Map()); await log(`success key=${k} chars=${answer.length} artifacts=${artifacts.length}`); resolve({ answer, artifacts }); }); c.stdin.end(prompt(k, text, files), "utf8"); });
+  return new Promise((resolve, reject) => {
+    const c = spawn(command, args, { cwd: workdir, env: childEnv, windowsHide: true, stdio: ["pipe", "pipe", "pipe"] });
+    let err = "";
+    // Codex --json can emit many events. Always drain stdout, or its pipe fills and stalls the child.
+    let stdoutBytes = 0;
+    c.stdout.on("data", b => { stdoutBytes += b.length; });
+    c.stderr.on("data", b => { err = (err + String(b)).slice(-6000); });
+    const configuredTimeout = Number(env("CODEX_TIMEOUT_MS", "900000"));
+    const timeoutMs = Number.isFinite(configuredTimeout) && configuredTimeout > 0 ? configuredTimeout : 900000;
+    let timedOut = false;
+    const timer = setTimeout(() => {
+      timedOut = true;
+      void log(`timeout key=${k} after=${timeoutMs}ms`);
+      c.kill("SIGKILL");
+      reject(new Error(`Codex 响应超时（${Math.round(timeoutMs / 60000)} 分钟）。请缩小任务或提高 CODEX_TIMEOUT_MS。`));
+    }, timeoutMs);
+    c.on("error", e => { clearTimeout(timer); reject(new Error(`无法启动 Codex：${e.message}`)); });
+    c.on("close", async code => {
+      try {
+      await log(`codex exit key=${k} code=${code} stdoutBytes=${stdoutBytes} timedOut=${timedOut}`);
+      clearTimeout(timer);
+      if (timedOut) return;
+      let answer = "";
+      try { answer = (await readFile(out, "utf8")).trim(); } catch {}
+      if (code !== 0) { await log(`error code=${code} stderr=${err.slice(-1500)}`); reject(new Error(err.trim().slice(-1000) || `Codex 退出码 ${code}`)); return; }
+      if (!answer) answer = "Codex 没有返回文本。";
+      const a = store[k] ??= [];
+      a.push({ role: "user", text, at: new Date().toISOString() }, { role: "assistant", text: answer, at: new Date().toISOString() });
+      if (a.length > 40) a.splice(0, a.length - 40);
+      await save();
+      const artifacts = await findNewArtifacts(workdir, before ?? new Map());
+      await log(`success key=${k} chars=${answer.length} artifacts=${artifacts.length}`);
+      resolve({ answer, artifacts });
+      } catch (e) { reject(e); }
+    });
+    c.stdin.on("error", e => { void log(`stdin error key=${k} ${e.message}`); });
+    c.stdin.end(prompt(k, text, files), "utf8");
+  });
 }
 export function splitReply(s: string) { const m = Math.max(500, Number(env("MAX_REPLY_CHARS", "3500"))), a: string[] = []; for (let i = 0; i < s.length; i += m) a.push(s.slice(i, i + m)); return a; }
 export async function snapshotArtifacts(workdir: string) { await mkdir(path.resolve(workdir), { recursive: true }); return listArtifacts(path.resolve(workdir)); }

src/index.ts

@@ -1,14 +1,78 @@
-import "dotenv/config";
-import { QQBot, mentionGate, messageFilter, rateLimiter, concurrencyGuard, accessPolicy, errorHandler } from "@tencent-connect/qqbot-nodejs";
-import { initStore, resetSession, askCodex, splitReply, downloadAttachments, snapshotArtifacts } from "./codex.js";
-for (const n of ["QQBOT_APP_ID", "QQBOT_APP_SECRET"]) if (!process.env[n]?.trim()) throw new Error(`缺少环境变量 ${n}`);
-const allowed = (process.env.QQ_ALLOWED_OPENIDS ?? "").split(",").map(x => x.trim()).filter(Boolean); if (!allowed.length) console.warn("QQ_ALLOWED_OPENIDS 为空，将拒绝所有消息。"); await initStore();
-const bot = new QQBot({ appId: process.env.QQBOT_APP_ID!, appSecret: process.env.QQBOT_APP_SECRET!, logger: console }); bot.use(errorHandler()); bot.use(messageFilter({ skipSelfEcho: true, dedup: { windowMs: 10000 } })); bot.use(accessPolicy({ c2c: { mode: "allowlist", allow: allowed }, group: { mode: "allowlist", allow: allowed } })); bot.use(mentionGate({ requireMentionInGroup: true })); bot.use(rateLimiter({ perSender: { max: 6, windowMs: 60000 }, global: { max: 30, windowMs: 60000 } })); bot.use(concurrencyGuard({ strategy: "queue" }));
-bot.on("ready", () => console.log("QQ 机器人已连接；等待消息。")); bot.on("error", e => console.error("QQ SDK error:", e));
-bot.on("message", async (_ctx, msg) => { const key = `${msg.kind}:${msg.replyTarget.targetId}`, input = msg.content.trim(); if (!input && !(msg.attachments?.length)) return; if (input === "/reset") { await resetSession(key); await bot.sendText(msg.replyTarget, "已清空当前会话。"); return; } if (input === "/status") { await bot.sendText(msg.replyTarget, `运行正常。会话：${key}`); return; }
-  const workdir = process.env.CODEX_WORKDIR || "/home/node/app/workspace"; await bot.sendText(msg.replyTarget, msg.attachments?.length ? "已收到图片/文件，Codex 正在处理…" : "已收到，Codex 正在处理…");
-  try { const files = await downloadAttachments(workdir, key, msg.attachments ?? []); const before = await snapshotArtifacts(workdir); const result = await askCodex(key, input || "请分析我发送的图片或文件，并给出结果。", files, before); for (const part of splitReply(result.answer)) await bot.sendText(msg.replyTarget, part); for (const file of result.artifacts) { try { await bot.sendFile(msg.replyTarget, { localPath: file }, { fileName: file.split(/[\\/]/).pop() }); } catch (e) { await bot.sendText(msg.replyTarget, `文件已生成但发送失败：${file.split(/[\\/]/).pop()}\n${e instanceof Error ? e.message : String(e)}`); } } } catch (e) { console.error(e); await bot.sendText(msg.replyTarget, `处理失败：${e instanceof Error ? e.message : String(e)}`); }
-}); await bot.start();
+import "dotenv/config";
+import { readFile } from "node:fs/promises";
+import path from "node:path";
+import { QQBot, mentionGate, messageFilter, rateLimiter, accessPolicy, errorHandler } from "@tencent-connect/qqbot-nodejs";
+import { initStore, resetSession, askCodex, splitReply, downloadAttachments, snapshotArtifacts } from "./codex.js";
 
+for (const n of ["QQBOT_APP_ID", "QQBOT_APP_SECRET"]) if (!process.env[n]?.trim()) throw new Error(`缺少环境变量 ${n}`);
+const allowed = (process.env.QQ_ALLOWED_OPENIDS ?? "").split(",").map(x => x.trim()).filter(Boolean);
+if (!allowed.length) console.warn("QQ_ALLOWED_OPENIDS 为空，将拒绝所有消息。");
+await initStore();
 
+const bot = new QQBot({ appId: process.env.QQBOT_APP_ID!, appSecret: process.env.QQBOT_APP_SECRET!, logger: console });
+bot.use(errorHandler());
+bot.use(messageFilter({ skipSelfEcho: true, dedup: { windowMs: 10000 } }));
+bot.use(accessPolicy({ c2c: { mode: "allowlist", allow: allowed }, group: { mode: "allowlist", allow: allowed } }));
+bot.use(mentionGate({ requireMentionInGroup: true }));
+bot.use(rateLimiter({ perSender: { max: 6, windowMs: 60000 }, global: { max: 30, windowMs: 60000 } }));
+// Do not queue /status behind a long-running Codex request.
+const active = new Map<string, number>();
+bot.on("ready", () => console.log("QQ 机器人已连接；等待消息。"));
+bot.on("error", e => console.error("QQ SDK error:", e));
 
+bot.on("message", async (_ctx, msg) => {
+  const key = `${msg.kind}:${msg.replyTarget.targetId}`;
+  const input = msg.content.trim();
+  if (!input && !(msg.attachments?.length)) return;
+  if (input === "/status") {
+    const started = active.get(key);
+    await bot.sendText(msg.replyTarget, started ? `任务仍在处理，已运行 ${Math.floor((Date.now() - started) / 60000)} 分钟。` : "机器人在线，当前会话没有运行中的任务。");
+    return;
+  }
+  if (input === "/diag") {
+    const files = ["bridge.log", "startup.log", "jupyter.log"];
+    const excerpts: string[] = [];
+    for (const file of files) {
+      try {
+        let lines = (await readFile(path.join(process.cwd(), "data", file), "utf8")).split("\n").filter(Boolean).slice(-8).join("\n");
+        for (const secret of [process.env.CODEX_API_KEY, process.env.QQBOT_APP_SECRET, process.env.JUPYTER_TOKEN, process.env.QQBOT_APP_ID]) if (secret) lines = lines.replaceAll(secret, "[REDACTED]");
+        excerpts.push(`${file}:\n${lines.slice(-900)}`);
+      } catch { excerpts.push(`${file}: 暂无日志`); }
+    }
+    await bot.sendText(msg.replyTarget, excerpts.join("\n\n").slice(-3000));
+    return;
+  }
+  if (input === "/reset") {
+    if (active.has(key)) { await bot.sendText(msg.replyTarget, "当前任务尚在运行；请等待任务结束后再重置。"); return; }
+    await resetSession(key);
+    await bot.sendText(msg.replyTarget, "已清空当前会话。");
+    return;
+  }
+  if (active.has(key)) { await bot.sendText(msg.replyTarget, "当前已有任务运行中。发送 /status 可查看耗时，请勿重复提交。"); return; }
+  active.set(key, Date.now());
+  let progress: ReturnType<typeof setInterval> | undefined;
+  try {
+    await bot.sendText(msg.replyTarget, msg.attachments?.length ? "已收到附件，Codex 正在处理…" : "已收到，Codex 正在处理…");
+    progress = setInterval(() => {
+      const mins = Math.floor((Date.now() - active.get(key)!) / 60000);
+      void bot.sendText(msg.replyTarget, `任务仍在运行（${mins} 分钟），完成或失败后会回复。`).catch(e => console.error("进度消息发送失败:", e));
+    }, 180000);
+    const workdir = process.env.CODEX_WORKDIR || "/home/node/app/workspace";
+    const files = await downloadAttachments(workdir, key, msg.attachments ?? []);
+    const before = await snapshotArtifacts(workdir);
+    const result = await askCodex(key, input || "请分析我发送的图片或文件，并给出结果。", files, before);
+    for (const part of splitReply(result.answer)) await bot.sendText(msg.replyTarget, part);
+    for (const file of result.artifacts) {
+      try { await bot.sendFile(msg.replyTarget, { localPath: file }, { fileName: file.split(/[\\/]/).pop() }); }
+      catch (e) { await bot.sendText(msg.replyTarget, `文件已生成但发送失败：${file.split(/[\\/]/).pop()}\n${e instanceof Error ? e.message : String(e)}`); }
+    }
+  } catch (e) {
+    console.error("任务失败:", e);
+    try { await bot.sendText(msg.replyTarget, `处理失败：${e instanceof Error ? e.message : String(e)}`); }
+    catch (sendError) { console.error("失败消息也无法发送:", sendError); }
+  } finally {
+    if (progress) clearInterval(progress);
+    active.delete(key);
+  }
+});
+await bot.start();