feat: initial arena-dashboard (FastAPI + dark UI + liveness probe)

app/probe.py

@@ -0,0 +1,86 @@
+"""
+Cron-скрипт: liveness probe всех моделей из DB.
+Запускается внутри контейнера по расписанию (cron / supervisord).
+"""
+from __future__ import annotations
+
+import json
+import os
+import sqlite3
+import sys
+import time
+import urllib.request
+from datetime import datetime, timezone
+
+DB_PATH = os.environ.get("DB_PATH", "/data/arena.db")
+OMNIROUTE_URL = os.environ.get("OMNIROUTE_URL", "http://192.168.1.101:20128")
+OMNIROUTE_KEY = os.environ.get("OMNIROUTE_API_KEY", "")
+TIMEOUT = int(os.environ.get("CHECK_TIMEOUT", "30"))
+CONCURRENCY = int(os.environ.get("CHECK_CONCURRENCY", "5"))
+
+
+def check_model(model_id: str) -> tuple[bool, int | None]:
+    t0 = time.time()
+    try:
+        body = json.dumps({
+            "model": model_id,
+            "messages": [{"role": "user", "content": "2+2=?"}],
+            "max_tokens": 5,
+        }).encode()
+        req = urllib.request.Request(
+            f"{OMNIROUTE_URL}/v1/chat/completions",
+            data=body,
+            headers={
+                "Authorization": f"Bearer {OMNIROUTE_KEY}",
+                "Content-Type": "application/json",
+            },
+        )
+        with urllib.request.urlopen(req, timeout=TIMEOUT) as r:
+            resp = json.loads(r.read())
+            content = resp["choices"][0]["message"]["content"]
+            ok = bool(content.strip())
+    except Exception as e:
+        print(f"  ✗ {model_id}: {e}", file=sys.stderr)
+        ok = False
+    latency = int((time.time() - t0) * 1000) if ok else None
+    return ok, latency
+
+
+def run():
+    db = sqlite3.connect(DB_PATH, check_same_thread=False)
+    models = [r[0] for r in db.execute("SELECT id FROM models").fetchall()]
+    db.close()
+
+    print(f"[probe] checking {len(models)} models @ {datetime.now(timezone.utc).isoformat()}")
+
+    from concurrent.futures import ThreadPoolExecutor, as_completed
+
+    def do(mid):
+        ok, lat = check_model(mid)
+        return mid, ok, lat
+
+    db = sqlite3.connect(DB_PATH, check_same_thread=False)
+    now = datetime.now(timezone.utc).isoformat()
+
+    with ThreadPoolExecutor(max_workers=CONCURRENCY) as ex:
+        futs = {ex.submit(do, m): m for m in models}
+        done = ok_count = 0
+        for f in as_completed(futs):
+            mid, ok, lat = f.result()
+            db.execute(
+                "INSERT INTO checks (model_id, checked_at, ok, latency_ms) VALUES (?,?,?,?)",
+                (mid, now, int(ok), lat),
+            )
+            done += 1
+            if ok:
+                ok_count += 1
+            status = "✓" if ok else "✗"
+            print(f"  {status} {mid} ({lat}ms)" if lat else f"  {status} {mid}")
+
+    db.commit()
+    db.close()
+    print(f"[probe] done: {ok_count}/{done} ok")
+
+
+if __name__ == "__main__":
+    run()