[3.1-3.5] FAM-Edge 全链路 - API-Gateway/Video-Preprocessor/AI-Orchestrator/模型适配器(基类+Ollama+Gemini)/熔断器/JSON解析容错 + 配置

fam-edge/src/fam_edge/model_adapters/gemini_adapter.py

@@ -0,0 +1,156 @@
+"""
+GeminiAdapter - Google Gemini 云端模型适配器
+
+provider_name = "gemini"
+模型: gemini-1.5-flash
+健康检查: GET models API
+熔断器: 启用，连续 5 次失败 -> OPEN 15 分钟
+"""
+import os
+import base64
+import requests
+from typing import List, Optional
+
+from .base_adapter import BaseModelAdapter
+from .circuit_breaker import CircuitBreaker
+from ..logger import setup_logger
+
+logger = setup_logger('fam-edge.gemini_adapter')
+
+
+class GeminiAdapter(BaseModelAdapter):
+    """Gemini 云端 VLM 适配器"""
+
+    def __init__(self, config: dict):
+        super().__init__("gemini", config)
+        self.model_name = config.get('model_name', 'gemini-1.5-flash')
+        self.api_key = config.get('api_key', '')
+        self.timeout = config.get('timeout', 8)
+        cb_cfg = config.get('circuit_breaker', {})
+        self._cb = CircuitBreaker(
+            threshold=cb_cfg.get('threshold', 5),
+            cooldown=cb_cfg.get('cooldown', 900),
+            enabled=cb_cfg.get('enabled', True)  # 云端默认启用
+        )
+        self._base_url = "https://generativelanguage.googleapis.com/v1beta"
+
+    def health_check(self) -> bool:
+        """GET models API，检查可用性"""
+        if not self.api_key:
+            logger.warning("Gemini API Key 未配置，健康检查失败")
+            return False
+        try:
+            resp = requests.get(
+                f"{self._base_url}/models?key={self.api_key}",
+                timeout=10
+            )
+            if resp.status_code == 200:
+                models = resp.json().get('models', [])
+                model_names = [m.get('name', '') for m in models]
+                has_model = any(self.model_name in name for name in model_names)
+                if has_model:
+                    logger.info(f"Gemini 健康检查通过: 模型 {self.model_name} 可用")
+                    return True
+                else:
+                    logger.warning(f"Gemini 健康检查失败: 模型 {self.model_name} 未找到")
+                    return False
+            return False
+        except Exception as e:
+            logger.error(f"Gemini 健康检查异常: {e}")
+            return False
+
+    def analyze_frames(self, frame_paths: List[str],
+                       frame_timestamps: List[str],
+                       known_members_context: str) -> Optional[str]:
+        """调用 Gemini 视觉分析"""
+        if self._cb.is_open():
+            logger.warning("Gemini 熔断器 OPEN，跳过调用")
+            return None
+
+        if not self.api_key:
+            logger.warning("Gemini API Key 未配置，跳过调用")
+            return None
+
+        # 构建 Prompt
+        n = len(frame_paths)
+        prompt = self._build_visual_prompt(n, frame_timestamps, known_members_context)
+
+        # 构建 inline_data
+        parts = [{"text": prompt}]
+        for path in frame_paths:
+            try:
+                with open(path, 'rb') as f:
+                    img_data = base64.b64encode(f.read()).decode('utf-8')
+                parts.append({
+                    "inline_data": {
+                        "mime_type": "image/jpeg",
+                        "data": img_data
+                    }
+                })
+            except Exception as e:
+                logger.error(f"读取图片失败 {path}: {e}")
+
+        try:
+            resp = requests.post(
+                f"{self._base_url}/models/{self.model_name}:generateContent?key={self.api_key}",
+                json={
+                    "contents": [{"parts": parts}],
+                    "generationConfig": {"temperature": 0.2, "topP": 0.8}
+                },
+                timeout=self.timeout
+            )
+
+            if resp.status_code == 200:
+                data = resp.json()
+                candidates = data.get('candidates', [])
+                if candidates:
+                    output = candidates[0].get('content', {}).get('parts', [{}])[0].get('text', '')
+                    self._cb.record_success()
+                    logger.info(f"Gemini 视觉分析完成，输出长度={len(output)}")
+                    return output
+                else:
+                    logger.warning("Gemini 返回空 candidates")
+                    self._cb.record_failure()
+                    return None
+            else:
+                logger.error(f"Gemini 调用失败: {resp.status_code} {resp.text[:200]}")
+                self._cb.record_failure()
+                return None
+
+        except requests.Timeout:
+            logger.warning(f"Gemini 调用超时 ({self.timeout}s)，降级跳过")
+            self._cb.record_failure()
+            return None
+        except Exception as e:
+            logger.error(f"Gemini 调用异常: {e}")
+            self._cb.record_failure()
+            return None
+
+    def get_timeout(self) -> int:
+        return self.timeout
+
+    def get_circuit_breaker(self) -> CircuitBreaker:
+        return self._cb
+
+    def _build_visual_prompt(self, n: int, timestamps: List[str], known_members: str) -> str:
+        ts_lines = '\n'.join(
+            f"[Image {i+1}] Time: {ts}" for i, ts in enumerate(timestamps)
+        )
+        return f"""You are a home surveillance video analysis assistant. Describe what you see in the following {n} images chronologically. Be objective.
+
+Timestamps:
+{ts_lines}
+
+For each image, report:
+1. People: count, clothing (color + type), visible actions
+2. Objects: toys, bottles, furniture, etc.
+3. Interactions: between people or people and objects
+
+Known family members (match by features, use real_name if matched, otherwise "PersonX"):
+{known_members or 'None'}
+
+Output format (plain text, one paragraph per image, keep timestamp markers):
+[Image 1] Time: {timestamps[0] if timestamps else ''}
+Description: ...
+
+Be concise and objective. Do not output JSON or markdown."""