53 lines
1.3 KiB
Python
53 lines
1.3 KiB
Python
"""
|
|
Lightweight event emitter for the GPU inference server.
|
|
|
|
Pushes debug logs to the same Redis stream as the pipeline orchestrator,
|
|
so GPU-side details (model load, VRAM, inference timing) appear in the
|
|
same log panel.
|
|
|
|
Only active when the request includes X-Job-Id header.
|
|
No dependency on the detect package.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import os
|
|
from datetime import datetime, timezone
|
|
|
|
import redis
|
|
|
|
REDIS_URL = os.environ.get("REDIS_URL", "redis://localhost:6379/0")
|
|
EVENTS_PREFIX = "detect_events"
|
|
|
|
_LEVEL_ORDER = {"DEBUG": 0, "INFO": 1, "WARN": 2, "ERROR": 3}
|
|
|
|
_redis_client = None
|
|
|
|
|
|
def _get_redis():
|
|
global _redis_client
|
|
if _redis_client is None:
|
|
_redis_client = redis.from_url(REDIS_URL, decode_responses=True)
|
|
return _redis_client
|
|
|
|
|
|
def log(job_id: str, stage: str, level: str, msg: str, log_level: str = "INFO"):
|
|
"""Push a log event to Redis if the level meets the threshold."""
|
|
if not job_id:
|
|
return
|
|
if _LEVEL_ORDER.get(level.upper(), 1) < _LEVEL_ORDER.get(log_level.upper(), 1):
|
|
return
|
|
|
|
r = _get_redis()
|
|
key = f"{EVENTS_PREFIX}:{job_id}"
|
|
event = json.dumps({
|
|
"event": "log",
|
|
"level": level,
|
|
"stage": stage,
|
|
"msg": msg,
|
|
"ts": datetime.now(timezone.utc).isoformat(),
|
|
})
|
|
r.rpush(key, event)
|
|
r.expire(key, 3600)
|