Use monotonic clock for interval measurement (#6211)

Signed-off-by: Lifu Huang <lifu.hlf@gmail.com>

python/sglang/srt/entrypoints/http_server.py

@@ -182,9 +182,9 @@ async def health_generate(request: Request) -> Response:
         async for _ in _global_state.tokenizer_manager.generate_request(gri, request):
             break
 
-    tic = time.time()
+    tic = time.perf_counter()
     task = asyncio.create_task(gen())
-    while time.time() < tic + HEALTH_CHECK_TIMEOUT:
+    while time.perf_counter() < tic + HEALTH_CHECK_TIMEOUT:
         await asyncio.sleep(1)
         if _global_state.tokenizer_manager.last_receive_tstamp > tic:
             task.cancel()

python/sglang/srt/entrypoints/http_server_engine.py

@@ -24,10 +24,10 @@ def launch_server_process(server_args: ServerArgs) -> multiprocessing.Process:
 
     base_url = server_args.url()
     timeout = 300.0  # Increased timeout to 5 minutes for downloading large models
-    start_time = time.time()
+    start_time = time.perf_counter()
 
     with requests.Session() as session:
-        while time.time() - start_time < timeout:
+        while time.perf_counter() - start_time < timeout:
             try:
                 headers = {
                     "Content-Type": "application/json; charset=utf-8",