Kris
ed2ee086de
主要变更： 1. 新增UIActionScheduler实现优先级UI任务调度，支持CRITICAL/HIGH/NORMAL/LOW四级优先级 2. 替换原有SendQueue为UIActionScheduler，统一所有UI操作的调度逻辑 3. 为截图、登录、重启等API添加调度器封装，支持超时、限流与fast-fail机制 4. 新增Prometheus监控指标，统计UI任务执行、等待耗时、超时、限流与快速失败情况 5. 为QrCapture添加命令执行超时保护，避免X11操作卡死 6. 扩展StatusResponse与状态接口，暴露调度器指标 7. 新增完整的单元测试覆盖调度器功能 8. 为看门狗注入调度器，实现kill前队列清空与快速失败窗口
362 lines
14 KiB
Python
362 lines
14 KiB
Python
from __future__ import annotations
|
||
|
||
import asyncio
|
||
import logging
|
||
import shutil
|
||
|
||
from fastapi import APIRouter
|
||
|
||
from woc_bridge.config import (
|
||
DIAGNOSTIC_ITEMS,
|
||
_DIAGNOSTIC_ITEM_BY_ID,
|
||
_require_db_reader,
|
||
_require_xdotool,
|
||
_require_send_queue,
|
||
_state,
|
||
)
|
||
from woc_bridge.db.coordinator import (
|
||
_auto_extract_with_lock,
|
||
_resolve_db_state,
|
||
with_db_retry,
|
||
)
|
||
from woc_bridge.messaging.ui_action_scheduler import Priority
|
||
from woc_bridge.models import (
|
||
BridgeError,
|
||
ConnectivityResponse,
|
||
DiagnosticItem,
|
||
DiagnosticRunResult,
|
||
)
|
||
|
||
logger = logging.getLogger("woc-bridge")
|
||
router = APIRouter()
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 路由：GET /api/diagnostic/connectivity
|
||
# ---------------------------------------------------------------------------
|
||
@router.get("/api/diagnostic/connectivity", response_model=ConnectivityResponse)
|
||
async def diagnostic_connectivity() -> ConnectivityResponse:
|
||
"""bridge 连通性检查（channels ProbeableAdapter 调用）。
|
||
|
||
MVP 阶段恒返回 reachable=true / latency_ms=0 / error=null。
|
||
真实实现可加：内部 health ping / 依赖项检查 / 平均响应时延。
|
||
|
||
Returns:
|
||
ConnectivityResponse：reachable / latency_ms / error
|
||
"""
|
||
return ConnectivityResponse(reachable=True, latency_ms=0, error=None)
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 路由：GET /api/diagnostic/items
|
||
# ---------------------------------------------------------------------------
|
||
@router.get("/api/diagnostic/items")
|
||
async def diagnostic_items() -> dict:
|
||
"""返回诊断项定义列表。
|
||
|
||
前端「诊断」页面用本接口渲染可执行的检查项卡片，每张卡片含：
|
||
- check_id（POST /api/diagnostic/run/{check_id} 的路径参数）
|
||
- name（卡片标题）/ description（卡片副标题）
|
||
- severity（critical/error/warn/info，决定卡片颜色与是否阻断）
|
||
- auto_repairable（true 时显示「自动修复」按钮，调 autofix 接口）
|
||
|
||
Returns:
|
||
{"items": [DiagnosticItem, ...]}（共 4 项：wechat_running /
|
||
login_state / db_accessible / xdotool_available）
|
||
"""
|
||
return {
|
||
"items": [DiagnosticItem(**item) for item in DIAGNOSTIC_ITEMS],
|
||
}
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 路由：POST /api/diagnostic/run/{check_id}
|
||
# ---------------------------------------------------------------------------
|
||
@router.post("/api/diagnostic/run/{check_id}", response_model=DiagnosticRunResult)
|
||
@with_db_retry
|
||
async def diagnostic_run(check_id: str) -> DiagnosticRunResult:
|
||
"""执行单个诊断项检查（channels DoctorAdapter 调用）。
|
||
|
||
支持的 check_id：
|
||
- wechat_running：pgrep -x wechat 检测进程，auto_repairable=true
|
||
- login_state：detect_login_state()，区分 not_running/not_logged_in/logged_in
|
||
- db_accessible：check_db_status() 区分 ok/not_found/encrypted/unreadable，
|
||
encrypted 时返回 repair_plan 提示独立专项解决
|
||
- xdotool_available：shutil.which('xdotool') 检测可执行文件
|
||
|
||
Args:
|
||
check_id: 路径参数，诊断项 ID（见 /api/diagnostic/items 返回）
|
||
|
||
Returns:
|
||
DiagnosticRunResult：含 passed / severity / message / auto_repairable
|
||
/ repair_plan（仅 db_accessible=encrypted/unreadable 时非空）
|
||
|
||
Raises:
|
||
BridgeError(INVALID_PARAMS): 未知 check_id（HTTP 400）
|
||
|
||
Notes:
|
||
- 不抛业务错误的诊断项：即使 passed=false 也返回 200 + 详细 message，
|
||
让调用方按字段判定后续动作（如显示修复建议、调 autofix）
|
||
- repair_plan 字段仅在 db_accessible 失败时给出建议，其他项为 None
|
||
"""
|
||
item = _DIAGNOSTIC_ITEM_BY_ID.get(check_id)
|
||
if item is None:
|
||
raise BridgeError(
|
||
code="INVALID_PARAMS",
|
||
message=f"未知诊断项: {check_id}",
|
||
)
|
||
|
||
xdotool = _require_xdotool()
|
||
db_reader = _require_db_reader()
|
||
|
||
if check_id == "wechat_running":
|
||
passed = await xdotool.is_wechat_running()
|
||
message = "微信进程运行中" if passed else "微信进程未运行"
|
||
elif check_id == "login_state":
|
||
state = await xdotool.detect_login_state()
|
||
passed = state == "logged_in"
|
||
message = f"登录态: {state}"
|
||
elif check_id == "db_accessible":
|
||
# 统一走 _resolve_db_state（含 salt/key/pid/验证逻辑）
|
||
db_state = await _resolve_db_state()
|
||
if db_state.db_accessible:
|
||
source = _state.key_cache.source or "unknown"
|
||
msg = "微信 DB 可读" if db_state.status == "ok" \
|
||
else f"DB 加密但已具备解密能力（key 来源: {source}）"
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=True,
|
||
severity=item["severity"],
|
||
message=msg,
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan=None,
|
||
)
|
||
if db_state.status == "not_found":
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message="未在 /config 下找到微信消息 DB",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan=None,
|
||
)
|
||
if db_state.status == "unreadable":
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message="DB 文件存在但不可读（权限问题）",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan="检查 /config/xwechat_files 下 DB 文件权限",
|
||
)
|
||
# need_init / key_invalid 且无有效 key
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message="DB 已加密（SQLCipher），需密钥才能读取",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan="通过 POST /api/db/decrypt 注入 64 位 hex 密钥，或设置 WOC_DB_KEY 环境变量，或启用 WOC_DB_KEY_AUTO_EXTRACT=true 自动提取",
|
||
)
|
||
elif check_id == "xdotool_available":
|
||
path = shutil.which("xdotool")
|
||
passed = path is not None
|
||
message = f"xdotool 路径: {path}" if passed else "xdotool 不可用"
|
||
else: # 理论不可达
|
||
raise BridgeError(
|
||
code="INVALID_PARAMS",
|
||
message=f"未知诊断项: {check_id}",
|
||
)
|
||
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=passed,
|
||
severity=item["severity"],
|
||
message=message,
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan=None,
|
||
)
|
||
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# 路由：POST /api/diagnostic/autofix/{check_id}
|
||
# ---------------------------------------------------------------------------
|
||
@router.post("/api/diagnostic/autofix/{check_id}", response_model=DiagnosticRunResult)
|
||
@with_db_retry
|
||
async def diagnostic_autofix(check_id: str) -> DiagnosticRunResult:
|
||
"""尝试自动修复诊断项（channels DoctorAdapter 调用）。
|
||
|
||
仅 auto_repairable=true 的项可调，否则抛 INVALID_PARAMS。
|
||
当前可修复项：
|
||
- wechat_running：pkill -x wechat 杀进程 → 等 2s 检查 autostart 是否拉起
|
||
→ 未拉起则 bridge 显式启动（start_wechat）作为兜底
|
||
- xdotool_available：不可修复，返回 passed=false + "请重建容器"
|
||
（xdotool 缺失说明容器构建异常，bridge 内无法修复，只能重建）
|
||
- db_accessible：加密 DB 时尝试自动提取 key（P1 能力），失败返回 repair_plan
|
||
|
||
Args:
|
||
check_id: 路径参数，诊断项 ID
|
||
|
||
Returns:
|
||
DiagnosticRunResult：含 passed / message（描述修复结果或失败原因）
|
||
|
||
Raises:
|
||
BridgeError(INVALID_PARAMS): 未知 check_id 或该项不可自动修复（HTTP 400）
|
||
|
||
Notes:
|
||
- wechat_running 修复流程：
|
||
1. pkill 杀旧进程
|
||
2. 等 2s 让 autostart watchdog 拉起
|
||
3. 若 autostart 已拉起 → 返回 passed=true + 新 PID
|
||
4. 若 autostart 未拉起 → bridge 显式启动（start_wechat）
|
||
5. 显式启动成功 → passed=true + 新 PID
|
||
6. 显式启动失败 → passed=false + repair_plan 提示
|
||
- 调用方收到 passed=true 后应间隔 3-5s 调 /api/diagnostic/run/wechat_running
|
||
确认实际状态（进程可能在初始化中，窗口尚未出现）
|
||
"""
|
||
item = _DIAGNOSTIC_ITEM_BY_ID.get(check_id)
|
||
if item is None:
|
||
raise BridgeError(
|
||
code="INVALID_PARAMS",
|
||
message=f"未知诊断项: {check_id}",
|
||
)
|
||
if not item["auto_repairable"]:
|
||
raise BridgeError(
|
||
code="INVALID_PARAMS",
|
||
message="该项不可自动修复",
|
||
)
|
||
|
||
if check_id == "wechat_running":
|
||
xdotool = _require_xdotool()
|
||
scheduler = _require_send_queue()
|
||
|
||
# 启动 fast-fail 窗口：覆盖 autostart 拉起新进程的 ~5-10s 期间，
|
||
# 避免队列内积压任务对新窗口执行 xdotool 失败形成风暴
|
||
if hasattr(scheduler, "mark_wechat_dead"):
|
||
try:
|
||
scheduler.mark_wechat_dead(15.0)
|
||
except Exception as exc:
|
||
logger.warning("diagnostic/autofix: mark_wechat_dead failed: %s", exc)
|
||
|
||
async def _autofix_flow():
|
||
# 1. pkill 旧进程（不强制 SIGKILL，给微信优雅退出机会）
|
||
try:
|
||
proc = await asyncio.create_subprocess_exec(
|
||
"pkill", "-x", "wechat",
|
||
stdout=asyncio.subprocess.DEVNULL,
|
||
stderr=asyncio.subprocess.DEVNULL,
|
||
)
|
||
await proc.wait()
|
||
except Exception:
|
||
pass
|
||
|
||
# 2. 等 2s 让 autostart watchdog 拉起
|
||
await asyncio.sleep(2)
|
||
|
||
# 3. 检查 autostart 是否已拉起
|
||
new_pid = await xdotool.check_wechat_pid()
|
||
|
||
# 4. autostart 未拉起，bridge 显式启动兜底
|
||
if new_pid is None:
|
||
new_pid = await xdotool.start_wechat(timeout_sec=10)
|
||
|
||
return new_pid
|
||
|
||
try:
|
||
if hasattr(scheduler, "execute"):
|
||
new_pid = await scheduler.execute(
|
||
_autofix_flow,
|
||
priority=Priority.CRITICAL,
|
||
wait_timeout_ms=30000,
|
||
trace_id="autofix_wechat_running",
|
||
)
|
||
else:
|
||
new_pid = await _autofix_flow()
|
||
except Exception as exc:
|
||
logger.warning("diagnostic/autofix: wechat_running autofix failed: %s", exc)
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message=f"autofix 执行失败: {exc}",
|
||
auto_repairable=True,
|
||
repair_plan="检查 X 会话是否正常（Xvnc/openbox），或通过 VNC 桌面手动启动微信",
|
||
)
|
||
|
||
# 5. 返回结果
|
||
if new_pid is not None:
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=True,
|
||
severity=item["severity"],
|
||
message=f"微信进程已恢复，PID={new_pid}",
|
||
auto_repairable=True,
|
||
repair_plan=None,
|
||
)
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message="autostart 未拉起且 bridge 显式启动失败",
|
||
auto_repairable=True,
|
||
repair_plan="检查 X 会话是否正常（Xvnc/openbox），或通过 VNC 桌面手动启动微信",
|
||
)
|
||
elif check_id == "db_accessible":
|
||
db_reader = _require_db_reader()
|
||
# 检查 DB 状态
|
||
status = await asyncio.to_thread(db_reader.check_db_status)
|
||
if status == "ok":
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=True,
|
||
severity=item["severity"],
|
||
message="DB 可读，无需修复",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan=None,
|
||
)
|
||
if status in ("need_init", "key_invalid"):
|
||
# 尝试自动提取 key（阻塞等待结果，复用公共加锁路径）
|
||
extracted = await _auto_extract_with_lock()
|
||
if extracted:
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=True,
|
||
severity=item["severity"],
|
||
message="自动提取 DB 密钥成功，DB 已可解密读取",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan=None,
|
||
)
|
||
else:
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message="自动提取 DB 密钥失败",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan="通过 POST /api/db/decrypt 手动注入 64 位 hex 密钥，或设置 WOC_DB_KEY 环境变量",
|
||
)
|
||
# not_found / unreadable 不可自动修复
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message=f"DB 状态: {status}，无法自动修复",
|
||
auto_repairable=item["auto_repairable"],
|
||
repair_plan="检查 /config/xwechat_files 下 DB 文件是否存在且权限正确",
|
||
)
|
||
elif check_id == "xdotool_available":
|
||
return DiagnosticRunResult(
|
||
check_id=check_id,
|
||
passed=False,
|
||
severity=item["severity"],
|
||
message="xdotool 不可用，请重建容器",
|
||
auto_repairable=True,
|
||
repair_plan=None,
|
||
)
|
||
|
||
# 理论不可达
|
||
raise BridgeError(
|
||
code="INVALID_PARAMS",
|
||
message=f"未知诊断项: {check_id}",
|
||
)
|