fix: prevent browser process leaks and improve robustness

- fix(browser): wrap context lifecycle in try/finally to ensure browser
  closes on exceptions and KeyboardInterrupt (login, sniffer, generator)
- fix(browser): replace time.sleep with Playwright native waits
  (wait_for, wait_for_timeout) for more reliable element interaction
- fix(browser): use parameterized page.evaluate instead of f-string
  JS injection in generator polling
- fix(api): add retry logic in wait_for_completion to survive transient
  network errors
- fix(config): add prepare_profile_dir to copy profile to temp dir,
  preventing Chromium SingletonLock conflicts when tools run concurrently
- fix(sniffer): stream API logs to tempfile instead of unbounded memory
  list to avoid OOM
- fix(api): specify encoding='utf-8' when loading cookies from file

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

hunyuan3dweb/api.py

@@ -190,8 +190,15 @@ class Hunyuan3DAPI:
             完成的创作信息
         """
         start_time = time.time()
+        last_error = None
         while time.time() - start_time < timeout:
+            try:
                 status = self.get_generation_status(creation_id)
+            except Exception as e:
+                last_error = e
+                print(f"查询状态失败: {e}，将在 {poll_interval} 秒后重试...")
+                time.sleep(poll_interval)
+                continue
 
             # Handle both wrapped {code, data} and flat response formats
             if "code" in status:
@@ -209,6 +216,8 @@ class Hunyuan3DAPI:
 
             time.sleep(poll_interval)
 
+        if last_error:
+            raise TimeoutError(f"Generation timeout after {timeout} seconds. Last error: {last_error}")
         raise TimeoutError(f"Generation timeout after {timeout} seconds")
 
 
@@ -217,7 +226,7 @@ from .config import get_cookie_path
 def load_cookies_from_file(filepath: Optional[str] = None) -> str:
     """从文件加载Cookie，默认读取用户配置目录的 cookies.txt"""
     path = filepath or str(get_cookie_path())
-    with open(path, 'r') as f:
+    with open(path, 'r', encoding='utf-8') as f:
         return f.read().strip()
 
 

hunyuan3dweb/api_complete.py

@@ -411,7 +411,7 @@ from .config import get_cookie_path
 def load_cookies_from_file(filepath: Optional[str] = None) -> str:
     """从文件加载Cookie，默认读取用户配置目录的 cookies.txt"""
     path = filepath or str(get_cookie_path())
-    with open(path, 'r') as f:
+    with open(path, 'r', encoding='utf-8') as f:
         return f.read().strip()
 
 

hunyuan3dweb/browser/generator.py

@@ -4,8 +4,12 @@
 使用 cloakbrowser 在浏览器内完成所有操作（包括签名生成）
 """
 
+import contextlib
 import json
 import os
+import shutil
+import sys
+import tempfile
 import time
 from datetime import datetime
 from cloakbrowser import launch_persistent_context
@@ -15,6 +19,36 @@ from ..config import get_profile_dir
 PROFILE_DIR = str(get_profile_dir())
 
 
+@contextlib.contextmanager
+def _temp_persistent_context(headless=True):
+    """
+    将标准 profile 复制到临时目录后启动持久化上下文，
+    避免与 login 进程发生 Chromium SingletonLock 冲突。
+    """
+    standard = str(get_profile_dir())
+    if not os.path.exists(standard):
+        raise FileNotFoundError(f"未找到登录状态目录: {standard}")
+
+    temp_root = tempfile.mkdtemp(prefix="hunyuan3dweb-profile-")
+    temp_profile = os.path.join(temp_root, "profile")
+    shutil.copytree(standard, temp_profile, dirs_exist_ok=True)
+
+    context = None
+    try:
+        context = launch_persistent_context(temp_profile, headless=headless)
+        yield context
+    finally:
+        if context is not None:
+            try:
+                context.close()
+            except Exception:
+                pass
+        try:
+            shutil.rmtree(temp_root)
+        except Exception:
+            pass
+
+
 def generate_3d(image_path, wait_for_complete=False, timeout=300):
     """
     上传图片并触发生成3D模型
@@ -30,10 +64,7 @@ def generate_3d(image_path, wait_for_complete=False, timeout=300):
     if not os.path.exists(image_path):
         raise FileNotFoundError(f"图片不存在: {image_path}")
 
-    if not os.path.exists(PROFILE_DIR):
+    with _temp_persistent_context(headless=True) as context:
-        raise FileNotFoundError(f"未找到登录状态目录: {PROFILE_DIR}")
-
-    context = launch_persistent_context(PROFILE_DIR, headless=True)
         page = context.new_page()
 
         # 用于存储结果
@@ -63,26 +94,25 @@ def generate_3d(image_path, wait_for_complete=False, timeout=300):
 
         page.on("response", handle_response)
 
-    try:
         print("[1/6] 打开首页...")
-        page.goto("https://3d.hunyuan.tencent.com/")
+        page.goto("https://3d.hunyuan.tencent.com/", timeout=60000)
         page.wait_for_timeout(3000)
 
         print("[2/6] 点击 AI创作...")
         page.locator("button").filter(has_text="AI创作").first.click()
-        page.wait_for_timeout(2000)
+        page.locator("text=图生3D").wait_for(state="visible", timeout=10000)
 
         print("[3/6] 点击 图生3D...")
         page.locator("text=图生3D").first.click()
-        page.wait_for_timeout(2000)
+        page.locator("text=单张图片").wait_for(state="visible", timeout=10000)
 
         print("[4/6] 点击 单张图片...")
         page.locator("text=单张图片").first.click()
-        page.wait_for_timeout(2000)
+        page.locator('input[type=file]').first.wait_for(state="visible", timeout=10000)
 
         print("[5/6] 上传图片...")
         page.locator('input[type=file]').first.set_input_files(image_path)
-        page.wait_for_timeout(3000)
+        page.locator("text=立即生成").wait_for(state="visible", timeout=10000)
 
         print("[6/6] 点击 立即生成...")
         page.locator("text=立即生成").first.click()
@@ -100,14 +130,14 @@ def generate_3d(image_path, wait_for_complete=False, timeout=300):
             last_status = None
             while time.time() - start < timeout:
                 # 轮询状态
-                status_result = page.evaluate(f'''
+                status_result = page.evaluate('''
-                async () => {{
+                async (creationsId) => {
-                  const resp = await fetch('https://3d.hunyuan.tencent.com/api/3d/creations/detail?creationsId={result["creationsId"]}', {{
+                  const resp = await fetch(`https://3d.hunyuan.tencent.com/api/3d/creations/detail?creationsId=${creationsId}`, {
-                    headers: {{'X-Source': 'web', 'Referer': 'https://3d.hunyuan.tencent.com/'}}
+                    headers: {'X-Source': 'web', 'Referer': 'https://3d.hunyuan.tencent.com/'}
-                  }});
+                  });
                   return await resp.json();
-                }}
+                }
-                ''')
+                ''', result["creationsId"])
 
                 status = status_result.get("status", "unknown")
                 progress = status_result.get("progress", 0)
@@ -134,17 +164,13 @@ def generate_3d(image_path, wait_for_complete=False, timeout=300):
 
         return result
 
-    finally:
-        context.close()
-
 
 def get_quota():
     """获取当前配额信息"""
-    context = launch_persistent_context(PROFILE_DIR, headless=True)
+    with _temp_persistent_context(headless=True) as context:
         page = context.new_page()
 
-    try:
+        page.goto("https://3d.hunyuan.tencent.com/", timeout=60000)
-        page.goto("https://3d.hunyuan.tencent.com/")
         page.wait_for_timeout(3000)
 
         result = page.evaluate('''
@@ -158,17 +184,14 @@ def get_quota():
         }
         ''')
         return result
-    finally:
-        context.close()
 
 
 def get_creations_list():
     """获取作品列表"""
-    context = launch_persistent_context(PROFILE_DIR, headless=True)
+    with _temp_persistent_context(headless=True) as context:
         page = context.new_page()
 
-    try:
+        page.goto("https://3d.hunyuan.tencent.com/", timeout=60000)
-        page.goto("https://3d.hunyuan.tencent.com/")
         page.wait_for_timeout(3000)
 
         result = page.evaluate('''
@@ -182,13 +205,9 @@ def get_creations_list():
         }
         ''')
         return result
-    finally:
-        context.close()
 
 
 if __name__ == "__main__":
-    import sys
-
     if len(sys.argv) < 2:
         print("用法:")
         print(f"  python {sys.argv[0]} quota          # 查询配额")

hunyuan3dweb/browser/login.py

@@ -47,12 +47,14 @@ def main():
     headless = not args.no_headless
 
     # 使用持久化上下文，cookie 和登录状态会保存到 PROFILE_DIR
+    context = None
+    try:
         context = launch_persistent_context(PROFILE_DIR, headless=headless)
         page = context.new_page()
 
         print("正在打开腾讯混元3D...")
-    page.goto("https://3d.hunyuan.tencent.com/")
+        page.goto("https://3d.hunyuan.tencent.com/", timeout=60000)
-    time.sleep(2)
+        page.wait_for_timeout(2000)
 
         # 检查是否已经是登录状态（没有登录按钮说明已登录）
         login_btn = page.locator("button").filter(has_text="登录").first
@@ -62,21 +64,19 @@ def main():
             _extract_and_save_cookies(context)
 
             if headless:
-            context.close()
                 return
         else:
             # 未登录，走登录流程
             login_btn.click()
-        time.sleep(1.5)
+            page.locator("text=邮箱").wait_for(state="visible", timeout=10000)
 
             # 切换到邮箱登录
             email_tab = page.locator("text=邮箱").first
             if email_tab.count() > 0:
                 email_tab.click()
-            time.sleep(1.5)
+                email_input.wait_for(state="visible", timeout=10000)
             else:
                 print("未找到邮箱登录选项")
-            context.close()
                 return
 
             # 定位元素
@@ -91,25 +91,22 @@ def main():
             email = input("请输入邮箱地址，按回车发送验证码: ").strip()
             if not email:
                 print("邮箱不能为空，退出")
-            context.close()
                 return
 
             email_input.fill(email)
-        time.sleep(0.5)
 
             # 勾选协议（如果未勾选）
             if checkbox.count() > 0:
                 is_checked = checkbox.evaluate("el => el.checked")
                 if not is_checked:
                     checkbox.evaluate("el => el.click()")
-                time.sleep(0.3)
+                    page.wait_for_timeout(300)
 
             if send_code_btn.count() > 0:
                 send_code_btn.click()
                 print("已点击发送验证码，请查收邮件...")
             else:
                 print("未找到发送验证码按钮")
-            context.close()
                 return
 
             # 第二次交互：输入验证码并登录
@@ -117,11 +114,9 @@ def main():
             code = input("请输入邮箱验证码，按回车登录: ").strip()
             if not code:
                 print("验证码不能为空，退出")
-            context.close()
                 return
 
             code_input.fill(code)
-        time.sleep(0.5)
 
             # 点击登录
             if login_submit_btn.count() > 0:
@@ -129,7 +124,6 @@ def main():
                 print("已点击登录按钮，等待跳转...")
             else:
                 print("未找到登录提交按钮")
-            context.close()
                 return
 
             # 等待登录成功跳转
@@ -145,7 +139,12 @@ def main():
         if not headless:
             print("\n按 Enter 键关闭浏览器并退出...")
             input()
+    finally:
+        if context is not None:
+            try:
                 context.close()
+            except Exception:
+                pass
 
 
 if __name__ == "__main__":

hunyuan3dweb/browser/sniffer.py

@@ -6,19 +6,22 @@
 
 import json
 import os
+import shutil
 import sys
+import tempfile
 import time
 from datetime import datetime
 from cloakbrowser import launch_persistent_context
 
-from ..config import get_profile_dir
+from ..config import get_profile_dir, prepare_profile_dir
 
 PROFILE_DIR = str(get_profile_dir())
 API_LOG_FILE = "./api_requests.log.json"
 
 
 def main():
-    logs = []
+    # 使用临时文件缓冲日志，避免长时间嗅探导致内存无限增长
+    log_buffer = tempfile.NamedTemporaryFile(mode="w+", suffix=".jsonl", delete=False, encoding="utf-8")
 
     def log_request(request):
         url = request.url
@@ -37,7 +40,8 @@ def main():
                     entry["body"] = request.post_data
                 except Exception:
                     pass
-            logs.append(entry)
+            log_buffer.write(json.dumps(entry, ensure_ascii=False) + "\n")
+            log_buffer.flush()
             print(f"[REQ] {request.method} {url}")
 
     def log_response(response):
@@ -62,7 +66,8 @@ def main():
                     entry["body_preview"] = text[:500]
             except Exception as e:
                 entry["body_error"] = str(e)
-            logs.append(entry)
+            log_buffer.write(json.dumps(entry, ensure_ascii=False) + "\n")
+            log_buffer.flush()
             print(f"[RES] {response.status} {url}")
 
     if not os.path.exists(PROFILE_DIR):
@@ -70,7 +75,11 @@ def main():
         print("请先运行 hunyuan3dweb-login 完成登录")
         sys.exit(1)
 
-    context = launch_persistent_context(PROFILE_DIR, headless=False)
+    profile_dir, temp_dir = prepare_profile_dir()
+
+    context = None
+    try:
+        context = launch_persistent_context(profile_dir, headless=False)
         page = context.new_page()
 
         page.on("request", log_request)
@@ -91,6 +100,26 @@ def main():
         print("所有 API 请求会实时打印在终端中")
         print("按 Enter 停止捕获并保存结果...\n")
         input()
+    finally:
+        if context is not None:
+            try:
+                context.close()
+            except Exception:
+                pass
+        if temp_dir is not None:
+            try:
+                shutil.rmtree(temp_dir)
+            except Exception:
+                pass
+
+    # 从临时文件读取并保存为 JSON 数组
+    try:
+        log_buffer.flush()
+        log_buffer.seek(0)
+        logs = [json.loads(line) for line in log_buffer if line.strip()]
+    finally:
+        log_buffer.close()
+        os.unlink(log_buffer.name)
 
     # 保存日志
     with open(API_LOG_FILE, "w", encoding="utf-8") as f:
@@ -99,8 +128,6 @@ def main():
     print(f"\n共捕获 {len([l for l in logs if l['type'] == 'request'])} 个请求")
     print(f"结果已保存到: {os.path.abspath(API_LOG_FILE)}")
 
-    context.close()
-
 
 if __name__ == "__main__":
     try:

hunyuan3dweb/config.py

@@ -1,5 +1,8 @@
 import os
+import shutil
+import tempfile
 from pathlib import Path
+from typing import Optional, Tuple
 
 
 def get_config_dir() -> Path:
@@ -21,3 +24,21 @@ def get_profile_dir() -> Path:
     path = get_config_dir() / "profile"
     path.mkdir(parents=True, exist_ok=True)
     return path
+
+
+def prepare_profile_dir() -> Tuple[str, Optional[str]]:
+    """
+    将标准 profile 复制到临时目录，避免 Chromium 锁冲突。
+
+    Returns:
+        (实际使用的 profile 路径, 临时目录路径或 None)。
+        如果使用了临时副本，调用者应在完成后用 shutil.rmtree 删除。
+    """
+    standard = str(get_profile_dir())
+    if not os.path.exists(standard):
+        return standard, None
+
+    temp_root = tempfile.mkdtemp(prefix="hunyuan3dweb-profile-")
+    temp_profile = os.path.join(temp_root, "profile")
+    shutil.copytree(standard, temp_profile, dirs_exist_ok=True)
+    return temp_profile, temp_root