ericwyuan/sentinel-home-ai
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(fam-edge): 云端模型调用统计 - 新增 model_calls 表(provider/model/时间/耗时/成功/失败原因)，gemini/nvidia 每次请求经 model_call_hook 记录并随 sync delta 下发；gemini 支持模型级 model_timeouts（lite 实测22.6s×4≈90s）

Browse Source
This commit is contained in:
ericwyuan
2026-08-21 12:21:49 +08:00
parent 1b0f62e5e9
commit b87b38b140
6 changed files with 90 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
fam-edge/src/fam_edge/model_adapters/base_adapter.py
View File

@@ -33,6 +33,21 @@ class BaseModelAdapter(ABC):
self.config = config
# 角色: vision=视觉分析, text=智能问答兜底(本地模型); 默认 vision
self.role = config.get('role', 'vision')
# 模型调用统计回调（由编排层注入）:
# hook(provider, model, started_at, duration_sec, success, error)
self.model_call_hook = None
def _emit_model_call(self, model: str, started_at: str,
duration_sec: float, success: bool,
error: str = ''):
"""上报一次模型调用统计（供前端展示成功/失败/耗时/失败原因）"""
hook = self.model_call_hook
if hook is None:
return
try:
hook(self.provider_name, model, started_at, duration_sec, success, error)
except Exception:
pass # 统计失败不影响主流程
def get_role(self) -> str:
"""返回适配器角色: 'vision' 或 'text'"""

28
fam-edge/src/fam_edge/model_adapters/gemini_adapter.py
View File

@@ -13,6 +13,7 @@ import os
import time
import json
import requests
from datetime import datetime
from typing import Dict, List, Optional
from .base_adapter import BaseModelAdapter
@@ -33,6 +34,10 @@ class GeminiAdapter(BaseModelAdapter):
m for m in config.get('fallback_models', []) if m and m != self.model_name]
self.api_key = self._resolve_key(config.get('api_key', ''))
self.timeout = config.get('timeout', 600)
# 模型级独立超时（最终值，不参与编排层 ×N 放大）: {model_name: seconds}
# 例: {"gemini-flash-lite-latest": 90}（按实测耗时 ×4 配置）
self.model_timeouts = {
str(k): int(v) for k, v in (config.get('model_timeouts') or {}).items()}
cb_cfg = config.get('circuit_breaker', {})
self._cb = CircuitBreaker(
threshold=cb_cfg.get('threshold', 3),
@@ -224,9 +229,13 @@ class GeminiAdapter(BaseModelAdapter):
{"text": prompt},
]
for model in self.model_chain:
# 模型级独立超时优先；未单独配置的用适配器默认（可能已被编排层 ×N 放大）
model_timeout = self.model_timeouts.get(model, self.timeout)
for attempt in range(2):
if attempt == 0:
logger.info(f"Gemini [{model}] 本轮请求超时 {self.timeout}s")
logger.info(f"Gemini [{model}] 本轮请求超时 {model_timeout}s")
started = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
t0 = time.time()
try:
resp = requests.post(
f"{self._base_url}/models/{model}:generateContent?key={self.api_key}",
@@ -234,14 +243,20 @@ class GeminiAdapter(BaseModelAdapter):
"generationConfig": {
"temperature": temperature,
"maxOutputTokens": max_tokens}},
timeout=self.timeout
timeout=model_timeout
)
except requests.Timeout:
logger.warning(f"Gemini [{model}] 视频请求超时 ({self.timeout}s)")
duration = time.time() - t0
self._emit_model_call(model, started, duration, False,
f"timeout({model_timeout}s)")
logger.warning(f"Gemini [{model}] 视频请求超时 ({model_timeout}s)")
break
except Exception as e:
duration = time.time() - t0
self._emit_model_call(model, started, duration, False, str(e))
logger.error(f"Gemini [{model}] 视频请求异常: {e}")
break
duration = time.time() - t0
if resp.status_code == 200:
cands = resp.json().get('candidates', [])
@@ -250,21 +265,28 @@ class GeminiAdapter(BaseModelAdapter):
for p in (cands[0].get('content', {}) if cands else {}).get('parts', [])
).strip() if cands else ''
if text:
self._emit_model_call(model, started, duration, True)
if model != self.model_name:
logger.info(f"Gemini 主模型不可用，由 fallback [{model}] 出结果")
return text
self._emit_model_call(model, started, duration, False, "empty_text")
logger.warning(f"Gemini [{model}] 返回空文本")
continue
detail = resp.text[:150].replace('\n', ' ')
if resp.status_code == 429:
self._emit_model_call(model, started, duration, False, "429_quota")
logger.warning(f"Gemini [{model}] 429 配额耗尽，切换下一模型")
break
if resp.status_code == 503:
if attempt == 0:
self._emit_model_call(model, started, duration, False, "503_overload_retry")
logger.warning(f"Gemini [{model}] 503 过载，3s 后重试")
time.sleep(3)
continue
self._emit_model_call(model, started, duration, False, "503_overload")
break
self._emit_model_call(model, started, duration, False,
f"http_{resp.status_code}")
logger.warning(f"Gemini [{model}] HTTP {resp.status_code}: {detail}")
break
return None

10
fam-edge/src/fam_edge/model_adapters/nvidia_adapter.py
View File

@@ -11,7 +11,9 @@ SDK: openai (NIM 兼容 OpenAI API 规范)
import os
import json
import re
import time
import requests
from datetime import datetime
from typing import Dict, List, Optional
from .base_adapter import BaseModelAdapter
@@ -144,6 +146,8 @@ class NvidiaVisionAdapter(BaseModelAdapter):
return None
prompt = self._build_video_prompt(known_members_context, event_start_time)
started = datetime.now().strftime('%Y-%m-%d %H:%M:%S')
t0 = time.time()
try:
resp = self._client.chat.completions.create(
model=self.model_name,
@@ -158,16 +162,20 @@ class NvidiaVisionAdapter(BaseModelAdapter):
extra_body={"media_io_kwargs": {"video": {"num_frames": 128}}},
timeout=self.timeout
)
duration = time.time() - t0
content = resp.choices[0].message.content
if not content:
self._emit_model_call(self.model_name, started, duration, False, "empty_content")
logger.warning("NVIDIA 视频分析返回空 content")
self._cb.record_failure()
return None
data = self._parse_json(content)
if not data or 'events' not in data:
self._emit_model_call(self.model_name, started, duration, False, "json_parse_failed")
logger.warning(f"NVIDIA 视频 JSON 解析失败: {content[:150]}")
self._cb.record_failure()
return None
self._emit_model_call(self.model_name, started, duration, True)
self._cb.record_success()
logger.info(f"NVIDIA 整视频分析完成，events={len(data.get('events', []))}")
return {
@@ -177,6 +185,8 @@ class NvidiaVisionAdapter(BaseModelAdapter):
"compute_provider": "nvidia",
}
except Exception as e:
duration = time.time() - t0
self._emit_model_call(self.model_name, started, duration, False, str(e))
self._cb.record_failure()
logger.warning(f"NVIDIA 视频分析异常: {e}")
return None