# Her saatte bir cache warmup (cron job)
import schedule
 
def warm_cache():
    """Dummy query gönder, cache'i ısıt."""
    client.messages.create(
        model="claude-sonnet-4-6",
        max_tokens=10,
        system=[
            {"type": "text", "text": KB, "cache_control": {"type": "ephemeral", "ttl": "1h"}},
        ],
        messages=[{"role": "user", "content": "Sistem hazır mı?"}],
    )
 
# Her 50 dakikada 1 (1h TTL ile uyumlu, eviction'dan önce refresh)
schedule.every(50).minutes.do(warm_cache)
 
while True:
    schedule.run_pending()
    time.sleep(60)

# Trafik düştüğünde dummy query at, TTL bitmesin
last_request_time = time.time()
CACHE_TTL_SECONDS = 4 * 60  # 5m TTL altında safety margin
 
def on_request():
    global last_request_time
    last_request_time = time.time()
 
def keep_warm_loop():
    """Background thread: TTL'e yaklaşırsa dummy query at."""
    while True:
        elapsed = time.time() - last_request_time
        if elapsed > CACHE_TTL_SECONDS:
            print("Cache soğuyor, refresh...")
            warm_cache()
            last_request_time = time.time()
        time.sleep(30)
 
threading.Thread(target=keep_warm_loop, daemon=True).start()

TENANT_IDS = ["tenant-a", "tenant-b", "tenant-c", ...]
 
def warm_all_tenants():
    for tid in TENANT_IDS:
        client.messages.create(
            ...
            system=[{"text": f"[T:{tid}]\n{KB}", "cache_control": {"ttl": "1h"}}],
            ...
        )
    print(f"Warmed {len(TENANT_IDS)} tenant caches.")
 
schedule.every(50).minutes.do(warm_all_tenants)