875a4f80d2
- model_adapter: measure_inference_latency() (max_tokens=1, 최소 부하) - backend_registry: - health latency baseline 학습 (초기 5회 max, 이후 EMA) - get_load_status(): inference 우선, health/queue 보조 - cache 30s + cooldown 10s + asyncio.Lock으로 자기증폭 루프 방지 - 조건: health > baseline*3 또는 사용자 명시 요청 시에만 ping - worker: - "system_status" 액션 — 사용자 상태 조회 시 force_measure - _build_system_status() 응답 빌더 (health/baseline/ping/queue) - route busy 안내를 get_load_status 기반으로 변경 Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
134 lines
4.9 KiB
Python
134 lines
4.9 KiB
Python
"""ModelAdapter — 범용 OpenAI-compat 모델 어댑터."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import logging
|
|
from collections.abc import AsyncGenerator
|
|
|
|
import httpx
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class ModelAdapter:
|
|
"""OpenAI-compatible /v1/chat/completions 백엔드 범용 어댑터.
|
|
Ollama, MLX 등 모두 동일 인터페이스로 호출."""
|
|
|
|
def __init__(
|
|
self,
|
|
name: str,
|
|
base_url: str,
|
|
model: str,
|
|
system_prompt: str,
|
|
temperature: float = 0.7,
|
|
timeout: float = 120.0,
|
|
max_tokens: int = 4096,
|
|
):
|
|
self.name = name
|
|
self.base_url = base_url
|
|
self.model = model
|
|
self.system_prompt = system_prompt
|
|
self.temperature = temperature
|
|
self.timeout = timeout
|
|
self.max_tokens = max_tokens
|
|
|
|
async def stream_chat(self, message: str, *, messages: list[dict] | None = None) -> AsyncGenerator[str, None]:
|
|
"""스트리밍 호출. content chunk를 yield. messages 직접 전달 가능."""
|
|
if messages is None:
|
|
messages = [
|
|
{"role": "system", "content": self.system_prompt},
|
|
{"role": "user", "content": message},
|
|
]
|
|
payload = {
|
|
"model": self.model,
|
|
"messages": messages,
|
|
"stream": True,
|
|
"temperature": self.temperature,
|
|
"max_tokens": self.max_tokens,
|
|
}
|
|
|
|
async with httpx.AsyncClient(timeout=self.timeout) as client:
|
|
try:
|
|
async with client.stream(
|
|
"POST",
|
|
f"{self.base_url}/v1/chat/completions",
|
|
json=payload,
|
|
) as resp:
|
|
if resp.status_code != 200:
|
|
body = await resp.aread()
|
|
logger.error("%s error %d: %s", self.name, resp.status_code, body.decode())
|
|
raise RuntimeError(f"{self.name} 응답 실패 ({resp.status_code})")
|
|
|
|
async for line in resp.aiter_lines():
|
|
line = line.strip()
|
|
if not line or not line.startswith("data: "):
|
|
continue
|
|
payload_str = line[len("data: "):]
|
|
if payload_str == "[DONE]":
|
|
return
|
|
try:
|
|
chunk = json.loads(payload_str)
|
|
delta = chunk.get("choices", [{}])[0].get("delta", {})
|
|
content = delta.get("content", "")
|
|
if content:
|
|
yield content
|
|
except (json.JSONDecodeError, IndexError, KeyError):
|
|
continue
|
|
|
|
except httpx.ConnectError:
|
|
logger.error("%s connection failed: %s", self.name, self.base_url)
|
|
raise
|
|
except httpx.ReadTimeout:
|
|
logger.error("%s read timeout", self.name)
|
|
raise
|
|
|
|
async def complete_chat(self, message: str, *, messages: list[dict] | None = None) -> str:
|
|
"""비스트리밍 호출. 전체 응답 텍스트 반환. messages 직접 전달 가능."""
|
|
if messages is None:
|
|
messages = [
|
|
{"role": "system", "content": self.system_prompt},
|
|
{"role": "user", "content": message},
|
|
]
|
|
payload = {
|
|
"model": self.model,
|
|
"messages": messages,
|
|
"stream": False,
|
|
"temperature": self.temperature,
|
|
"max_tokens": self.max_tokens,
|
|
}
|
|
|
|
async with httpx.AsyncClient(timeout=self.timeout) as client:
|
|
resp = await client.post(
|
|
f"{self.base_url}/v1/chat/completions",
|
|
json=payload,
|
|
)
|
|
resp.raise_for_status()
|
|
data = resp.json()
|
|
return data["choices"][0]["message"]["content"]
|
|
|
|
async def health_check(self) -> bool:
|
|
"""GET /v1/models — 5초 timeout."""
|
|
try:
|
|
async with httpx.AsyncClient(timeout=5.0) as client:
|
|
resp = await client.get(f"{self.base_url}/v1/models")
|
|
return resp.status_code < 500
|
|
except Exception:
|
|
return False
|
|
|
|
async def measure_inference_latency(self) -> float:
|
|
"""ping 메시지로 실제 inference latency 측정. max_tokens=1로 최소 부하.
|
|
반환: 밀리초. 실패 시 -1.0"""
|
|
import time as _time
|
|
original_max = self.max_tokens
|
|
self.max_tokens = 1
|
|
try:
|
|
start = _time.monotonic()
|
|
await self.complete_chat("ping")
|
|
return (_time.monotonic() - start) * 1000
|
|
except Exception:
|
|
logger.warning("inference latency measurement failed for %s", self.name)
|
|
return -1.0
|
|
finally:
|
|
self.max_tokens = original_max
|