from dotenv import load_dotenv
load_dotenv('/opt/rinet-gpu/.env.master')
# auto-added by patch_scrapers_with_dotenv.sh
"""Debug observability dashboard endpoint."""
import json, os, time
from pathlib import Path
from fastapi import APIRouter, Query
from fastapi.responses import JSONResponse, HTMLResponse, PlainTextResponse
from typing import Optional
router = APIRouter(prefix="/api/debug", tags=["debug"])
LOGDIR = Path("/var/log/pgz-sport-debug")
@router.get("/health")
def debug_health():
"""Quick service status."""
import subprocess
services = ['pgz-sport', 'pgz-debug-tail', 'pgz-auto-triage', 'nginx', 'redis-server']
status = {}
for s in services:
try:
r = subprocess.run(['systemctl', 'is-active', s], capture_output=True, text=True, timeout=2)
status[s] = r.stdout.strip()
except Exception as e:
status[s] = f"error:{e}"
# DB
db_status = "unknown"
try:
import psycopg2
with psycopg2.connect(f"host=10.10.0.2 port=6432 dbname=rinet_v3 user=rinet password={os.environ['DB_PASSWORD']}", connect_timeout=2) as conn:
with conn.cursor() as cur:
cur.execute("SELECT 1")
db_status = "ok"
except Exception as e:
db_status = f"error:{e}"
# Recent errors count
err_count = 0
if (LOGDIR / "errors.jsonl").exists():
with open(LOGDIR / "errors.jsonl") as f:
err_count = sum(1 for _ in f)
return {
"ts": time.time(),
"services": status,
"db": db_status,
"total_errors_logged": err_count,
"log_dir": str(LOGDIR),
}
@router.get("/errors")
def recent_errors(limit: int = Query(100, ge=1, le=1000)):
"""Last N errors from errors.jsonl."""
f = LOGDIR / "errors.jsonl"
if not f.exists():
return {"errors": [], "note": "errors.jsonl not yet created"}
lines = f.read_text(errors='ignore').strip().split('\n')[-limit:]
parsed = []
for line in lines:
try:
parsed.append(json.loads(line))
except:
continue
return {"errors": parsed, "count": len(parsed)}
@router.get("/decisions")
def triage_decisions(limit: int = Query(50, ge=1, le=500)):
"""Last N auto-triage decisions."""
f = LOGDIR / "triage_decisions.jsonl"
if not f.exists():
return {"decisions": [], "note": "no decisions yet"}
lines = f.read_text(errors='ignore').strip().split('\n')[-limit:]
parsed = []
for line in lines:
try:
parsed.append(json.loads(line))
except:
continue
return {"decisions": parsed, "count": len(parsed)}
@router.get("/stream")
def stream_tail(lines: int = Query(200, ge=10, le=2000)):
"""Last N lines of full stream.jsonl."""
f = LOGDIR / "stream.jsonl"
if not f.exists():
return {"stream": []}
raw = f.read_text(errors='ignore').strip().split('\n')[-lines:]
parsed = []
for line in raw:
try:
parsed.append(json.loads(line))
except:
continue
return {"stream": parsed}
@router.get("/dashboard", response_class=HTMLResponse)
def dashboard():
"""Live HTML dashboard."""
return """
PGŽ Debug Live
🩺 PGŽ Sport · Live Debug Dashboard refresh: 5s
Service Health
loading…
Auto-Triage Decisions
loading…
"""