Increase flush cache timeout in hicache CI (#21305)

python/sglang/test/test_utils.py

@@ -169,21 +169,27 @@ def download_image_with_retry(image_url: str, max_retries: int = 3) -> Image.Ima
 
 
 def flush_cache_with_retry(
-    base_url: str, retries: int = 5, interval: float = 2.0
+    base_url: str,
+    timeout: float = 30.0,
+    poll_interval: float = 0.5,
 ) -> bool:
-    """Flush device cache with retry.
+    """Flush device cache, polling until success or timeout.
 
-    The scheduler may still have in-flight HiCache async ops (GPU↔Host↔L3)
+    flush_cache only succeeds when the scheduler is fully idle, but
-    that prevent is_fully_idle() from returning True, so we retry.
+    HiCache async ops (write-through, backup) may still be in-flight
+    after a request completes.  We poll with a short interval so idle
+    is detected quickly, while the generous timeout accommodates slow
+    CI environments.
     """
-    for _ in range(retries):
+    deadline = time.time() + timeout
+    while time.time() < deadline:
         try:
             response = requests.post(f"{base_url}/flush_cache", timeout=10)
             if response.status_code == 200:
                 return True
         except requests.RequestException:
             pass
-        time.sleep(interval)
+        time.sleep(poll_interval)
     return False