MoFin/deploy/profile-scripts/self_repair.py

#!/usr/bin/env python3
"""self_repair.py — L3 LLM 修复循环（报备制）

读取 L1 functional_health.json + L2 hygiene_report.json 的失败项，
调用 LLM 诊断并**直接执行修复**（先斩后奏），然后记录日志并 XMPP 报备。

安全设计：
- LLM 只能从白名单动作中选择（不许任意执行代码）
- 每个模块每天最多自动修复 2 次（防修复循环）
- LLM 不可用时降级为规则默认动作（rerun_script）
- 所有动作记录 repair_log.jsonl

频率：每 30 分钟（cron）。
"""
import os, sys, json, subprocess, sqlite3, time
from datetime import datetime

sys.path.insert(0, '/home/hmo/MoFin')

FUNCTIONAL_REPORT = '/home/hmo/MoFin/gateway/logs/functional_health.json'
HYGIENE_REPORT = '/home/hmo/MoFin/gateway/logs/hygiene_report.json'
REPAIR_LOG = '/home/hmo/MoFin/gateway/logs/repair_log.jsonl'
REPAIR_STATE = '/home/hmo/MoFin/gateway/logs/repair_state.json'
GATEWAY = 'http://127.0.0.1:8643/v1/chat/completions'
SCRIPTS_DIR = '/home/hmo/.hermes/profiles/position-analyst/scripts'
MAX_REPAIRS_PER_MODULE_PER_DAY = 2

WHITELIST_ACTIONS = """
可执行的白名单动作（只能选其一，不许自创）：
1. {"action": "rerun_script", "script": "<脚本名>"} — 立即重跑指定 cron 脚本（限注册表内的脚本）
2. {"action": "restart_service", "service": "<服务名>"} — systemctl 重启（限: hermes-gateway-zhiwei, xmpp-zhiwei, mofin-dashboard）
3. {"action": "sync_links"} — 重建脚本硬链接（修复 cron 跑旧代码）
4. {"action": "switch_llm_key"} — 切换 LLM key（429 限流时）
5. {"action": "none", "reason": "<原因>"} — 判断为无需动作（如非交易时段的正常空闲）
"""


def log_repair(entry):
    os.makedirs(os.path.dirname(REPAIR_LOG), exist_ok=True)
    with open(REPAIR_LOG, 'a', encoding='utf-8') as f:
        f.write(json.dumps(entry, ensure_ascii=False) + '\n')


def load_state():
    try:
        return json.load(open(REPAIR_STATE))
    except Exception:
        return {}


def save_state(s):
    with open(REPAIR_STATE, 'w') as f:
        json.dump(s, f)


def repairs_today(state, module):
    today = datetime.now().strftime('%Y-%m-%d')
    return state.get(today, {}).get(module, 0)


def bump_repair(state, module):
    today = datetime.now().strftime('%Y-%m-%d')
    state.setdefault(today, {})
    state[today][module] = state[today].get(module, 0) + 1
    # 只保留最近 3 天
    for k in list(state.keys()):
        if k < (datetime.now().strftime('%Y-%m-%d'))[:8] + '17':
            del state[k]
    save_state(state)


def collect_failures():
    failures = []
    try:
        fh = json.load(open(FUNCTIONAL_REPORT))
        for c in fh.get('checks', []):
            if c.get('status') == 'fail':
                failures.append({
                    'source': 'functional', 'module': c['module'],
                    'function': c['function'], 'reason': c['reason'],
                    'repair_hint': c.get('repair'),
                })
    except Exception:
        pass
    try:
        hy = json.load(open(HYGIENE_REPORT))
        for i in hy.get('issues', []):
            failures.append({
                'source': 'hygiene', 'module': i.get('type'),
                'function': i.get('file') or i.get('job') or i.get('table', '?'),
                'reason': json.dumps(i, ensure_ascii=False)[:200],
                'repair_hint': {'action': 'llm_diagnose'},
            })
    except Exception:
        pass
    return failures


def run_action(action, module):
    """执行白名单动作，返回 (ok, detail)"""
    act = action.get('action')
    try:
        if act == 'rerun_script':
            script = action.get('script', '')
            # 只允许注册表内脚本
            if not script.endswith('.py') or '/' in script or '..' in script:
                return False, f'非法脚本名: {script}'
            path = os.path.join(SCRIPTS_DIR, script)
            if not os.path.exists(path):
                return False, f'脚本不存在: {script}'
            r = subprocess.run(['python3', path], capture_output=True, text=True,
                               timeout=600, cwd=SCRIPTS_DIR)
            tail = (r.stdout or r.stderr)[-150:]
            return r.returncode == 0, f'exit={r.returncode} {tail}'

        elif act == 'restart_service':
            svc = action.get('service', '')
            allowed = {'hermes-gateway-zhiwei': ['sudo', '-n', 'systemctl', 'restart', svc],
                       'xmpp-zhiwei': ['sudo', '-n', 'systemctl', 'restart', svc],
                       'mofin-dashboard': ['sudo', '-n', 'systemctl', 'restart', svc]}
            if svc not in allowed:
                return False, f'服务不在白名单: {svc}'
            subprocess.Popen(allowed[svc], stdout=subprocess.DEVNULL,
                             stderr=subprocess.DEVNULL, start_new_session=True)
            return True, f'{svc} 重启已触发(异步)'

        elif act == 'sync_links':
            r = subprocess.run(['bash', '/home/hmo/MoFin/deploy/profile-scripts/sync_profile_scripts.sh'],
                               capture_output=True, text=True, timeout=60)
            return True, r.stdout.strip()

        elif act == 'switch_llm_key':
            sys.path.insert(0, '/home/hmo/MoFin')
            from xmpp_logger import best_key, switch_key, KEY_TO_PROVIDER, current_provider
            bk = best_key()
            if not bk:
                return False, '无可用 key'
            target = KEY_TO_PROVIDER.get(bk['key_id'])
            if not target or target == current_provider():
                return False, f'已在最优 key ({bk["key_id"]})'
            r = switch_key(bk['key_id'])
            return r.get('switched', False), json.dumps(r, ensure_ascii=False)[:150]

        elif act == 'none':
            return True, f'无需动作: {action.get("reason", "")}'

        return False, f'未知动作: {act}'
    except subprocess.TimeoutExpired:
        return False, '动作超时(600s)'
    except Exception as e:
        return False, f'动作异常: {str(e)[:120]}'


def llm_diagnose(failure):
    """调 LLM 诊断并返回白名单动作。LLM 不可用 → 返回 None（走规则默认）"""
    hint = failure.get('repair_hint') or {}
    default_action = hint if hint.get('action') and hint['action'] != 'llm_diagnose' else None

    prompt = f"""你是 MoFin 系统的自动修复工程师。一个健康检查发现了功能异常，请诊断并给出一个修复动作。

【异常信息】
模块: {failure['module']}
功能: {failure['function']}
异常表现: {failure['reason']}
来源: {failure['source']}

【背景】
- 系统: Linux 246, MoFin 股票分析系统
- cron 脚本目录: {SCRIPTS_DIR}
- 数据库: /home/hmo/MoFin/data/mofin.db
- 当前时间: {datetime.now().strftime('%Y-%m-%d %H:%M')}（注意是否交易时段，非交易时段部分管道停跑属正常）

{WHITELIST_ACTIONS}

请只输出一个 JSON 动作（不要其他文字）。如果是非交易时段的正常空闲，选 none。"""

    try:
        payload = json.dumps({
            'model': 'deepseek-v4-flash',
            'messages': [{'role': 'user', 'content': prompt}],
            'max_tokens': 300, 'stream': False,
        }).encode()
        import urllib.request
        req = urllib.request.Request(GATEWAY, data=payload, headers={
            'Content-Type': 'application/json', 'Authorization': 'Bearer hermes123',
            'X-Hermes-Session-Id': 'self-repair'})
        resp = urllib.request.urlopen(req, timeout=120)
        text = json.loads(resp.read())['choices'][0]['message']['content']
        # 提取 JSON
        import re
        m = re.search(r'\{[^{}]*"action"[^{}]*\}', text)
        if m:
            action = json.loads(m.group(0))
            if action.get('action') in ('rerun_script', 'restart_service', 'sync_links',
                                        'switch_llm_key', 'none'):
                return action, text[:200]
    except Exception as e:
        print(f'  LLM 诊断不可用: {str(e)[:80]}')
    return (default_action, None) if default_action else (None, None)


def xmpp_report(lines):
    try:
        import urllib.request
        payload = json.dumps({'to': 'hmo@yoin.fun', 'body': '\n'.join(lines), 'type': 'chat'}).encode()
        req = urllib.request.Request('http://127.0.0.1:5805/', data=payload,
                                     headers={'Content-Type': 'application/json'})
        urllib.request.urlopen(req, timeout=5)
    except Exception as e:
        print(f'XMPP 失败: {e}')


def main():
    now = datetime.now()
    print('🔧 self_repair', now.strftime('%H:%M'))
    failures = collect_failures()
    if not failures:
        print('  ✅ 无失败项')
        return

    state = load_state()
    reports = []
    for f in failures:
        module = f['module']
        if repairs_today(state, module) >= MAX_REPAIRS_PER_MODULE_PER_DAY:
            print(f'  ⏭ {module}: 今日已修复 {MAX_REPAIRS_PER_MODULE_PER_DAY} 次，跳过（防循环）')
            continue

        print(f'  🔍 {module}: {f["reason"][:60]}')
        action, llm_note = llm_diagnose(f)
        if action is None:
            print(f'    无可用动作，跳过')
            log_repair({'ts': now.isoformat(), 'module': module, 'reason': f['reason'],
                        'action': 'skip_no_action', 'ok': None})
            continue

        ok, detail = run_action(action, module)
        bump_repair(state, module)
        entry = {
            'ts': now.isoformat(), 'module': module, 'function': f['function'],
            'reason': f['reason'], 'action': action, 'ok': ok, 'detail': detail[:300],
            'llm_note': llm_note,
        }
        log_repair(entry)
        icon = '✅' if ok else '❌'
        print(f'    {icon} {action.get("action")} -> {detail[:80]}')
        reports.append(f'{icon} [{module}] {action.get("action")}: {detail[:80]}')

    if reports:
        xmpp_report(['🔧 自检系统自动修复报备：', f'发现 {len(failures)} 个功能异常，已直接处理：'] + reports +
                    ['', '详情: gateway/logs/repair_log.jsonl'])


if __name__ == '__main__':
    main()