feat: requirements analysis, submission checks (stability stress test), AI inspection pipeline

2026-08-31 13:47:29 +08:00
parent a9bbe2f9b4
commit eaffeb58f5
17 changed files with 1218 additions and 8 deletions
@@ -0,0 +1,133 @@
+import json
+import logging
+
+import requests
+from django.conf import settings
+
+logger = logging.getLogger(__name__)
+
+ANALYZE_PROMPT = """\
+你是 OJ 检测平台的分析引擎。请根据下面的作业要求,分析出需要对该成员提交的项目做哪些检测。
+
+必须始终包含 ai_inspection(AI 检测)。根据作业要求酌情包含以下维度:
+- stability: 稳定度检测(压测)。requirements 用中文给出具体指标,例如:
+  基准并发 {stability_concurrency} 下错误率 ≤ 1%, 基准持续 {base_duration}s 后加压 50% 持续 {rampup_duration}s 崩溃率 ≤ 0.5%
+- functional: 功能点完善度检测。points 为需要核实的中文功能点列表(必须覆盖作业要求中的全部功能点)
+- performance: 性能检测。requirements 为中文指标,例如: 基准并发下 P95 延迟 ≤ 500ms
+- security: 安全检测(漏洞扫描),requirements 为中文说明
+
+输出格式: 只输出一个 JSON 对象,不要输出任何其他文字:
+{{"ai_inspection": true, "stability": {{"enabled": true, "requirements": "中文要求"}}, "functional": {{"enabled": true, "points": ["功能点1", "功能点2"]}}, "performance": {{"enabled": false, "requirements": ""}}, "security": {{"enabled": false, "requirements": ""}}}}
+
+作业要求如下:
+{content}
+"""
+
+
+def _validate_spec(spec: dict) -> dict:
+    if not isinstance(spec, dict):
+        raise ValueError("LLM 未返回 JSON 对象")
+    result: dict = {"ai_inspection": bool(spec.get("ai_inspection", True))}
+    for key in ("stability", "performance", "security"):
+        node = spec.get(key)
+        if isinstance(node, dict):
+            result[key] = {
+                "enabled": bool(node.get("enabled")),
+                "requirements": str(node.get("requirements", "")),
+            }
+        else:
+            result[key] = {"enabled": False, "requirements": ""}
+    func = spec.get("functional")
+    if isinstance(func, dict):
+        points = func.get("points")
+        result["functional"] = {
+            "enabled": bool(func.get("enabled")),
+            "points": [str(p) for p in points] if isinstance(points, list) else [],
+        }
+    else:
+        result["functional"] = {"enabled": True, "points": []}
+    return result
+
+
+def _extract_json(text: str) -> dict:
+    start, end = text.find("{"), text.rfind("}")
+    if start == -1 or end <= start:
+        raise ValueError("LLM 响应中未找到 JSON")
+    return json.loads(text[start:end + 1])
+
+
+def analyze_requirements(content: str) -> dict:
+    """调用 LLM 分析作业要求,返回检测维度计划。未配置 LLM 时返回默认计划。"""
+    key = settings.LLM_API_KEY
+    if not key or key.startswith("sk-placeholder") or not settings.LLM_MODEL:
+        logger.warning("LLM 未配置,使用默认检测计划")
+        return _default_spec()
+    url = settings.LLM_BASE_URL.rstrip("/") + "/chat/completions"
+    prompt = ANALYZE_PROMPT.format(
+        content=content[:8000],
+        stability_concurrency=settings.STABILITY_CONCURRENCY,
+        base_duration=settings.STABILITY_BASE_DURATION,
+        rampup_duration=settings.STABILITY_RAMPUP_DURATION,
+    )
+    resp = requests.post(
+        url,
+        headers={"Authorization": f"Bearer {key}", "Content-Type": "application/json"},
+        json={
+            "model": settings.LLM_MODEL,
+            "messages": [
+                {"role": "system", "content": "你是严格遵循 JSON 输出格式的 OJ 检测分析引擎。"},
+                {"role": "user", "content": prompt},
+            ],
+            "temperature": 0.1,
+        },
+        timeout=120,
+    )
+    resp.raise_for_status()
+    text = resp.json()["choices"][0]["message"]["content"]
+    spec = _extract_json(text)
+    return _validate_spec(spec)
+
+
+def _default_spec() -> dict:
+    return {
+        "ai_inspection": True,
+        "stability": {"enabled": True, "requirements": "默认稳定度要求: 基准并发下错误率≤1%, 加压后崩溃率≤0.5%"},
+        "functional": {"enabled": True, "points": []},
+        "performance": {"enabled": True, "requirements": "默认性能要求: 基准并发下 P95 延迟 ≤ 500ms"},
+        "security": {"enabled": False, "requirements": ""},
+    }
+
+
+def ai_summarize(project_dir: str, spec: dict) -> dict:
+    """AI 检测: 阅读项目关键文件,对照检测计划输出中文报告片段。"""
+    key = settings.LLM_API_KEY
+    if not key or key.startswith("sk-placeholder") or not settings.LLM_MODEL:
+        return {"summary": "LLM 未配置,跳过 AI 检测。请在 .env 中配置 LLM_API_KEY。", "issues": []}
+    from .code_reader import read_key_files
+    excerpt = read_key_files(project_dir, max_chars=12000)
+    prompt = (
+        "你是 OJ 平台的代码审查专家。下面是成员提交的项目关键代码摘录和检测计划(JSON):\n"
+        f"{json.dumps(spec, ensure_ascii=False)}\n\n"
+        f"代码摘录:\n{excerpt}\n\n"
+        "请输出一个 JSON 对象: {\"summary\": \"中文总体评价(200字内)\", "
+        "\"issues\": [{\"point\": \"检查点\", \"status\": \"pass|fail|warn\", \"detail\": \"中文说明\"}]}\n"
+        "对照检测计划逐项核查,重点关注: 功能点是否完善、稳定度是否可能达标、潜在风险。只输出 JSON。"
+    )
+    url = settings.LLM_BASE_URL.rstrip("/") + "/chat/completions"
+    resp = requests.post(
+        url,
+        headers={"Authorization": f"Bearer {key}", "Content-Type": "application/json"},
+        json={
+            "model": settings.LLM_MODEL,
+            "messages": [{"role": "user", "content": prompt}],
+            "temperature": 0.2,
+        },
+        timeout=180,
+    )
+    resp.raise_for_status()
+    text = resp.json()["choices"][0]["message"]["content"]
+    data = _extract_json(text)
+    return {
+        "summary": str(data.get("summary", "")),
+        "issues": data.get("issues", []) if isinstance(data.get("issues"), list) else [],
+    }