Crash the server when error or OOM happens (#514)

python/sglang/srt/utils.py

@@ -12,6 +12,7 @@ from io import BytesIO
 from typing import List, Optional
 
 import numpy as np
+import psutil
 import requests
 import rpyc
 import torch
@@ -441,6 +442,27 @@ def assert_pkg_version(pkg: str, min_version: str):
         )
 
 
+def kill_parent_process():
+    """Kill the parent process and all children of the parent process."""
+    current_process = psutil.Process()
+    parent_process = current_process.parent()
+    children = current_process.children(recursive=True)
+    for child in children:
+        if child.pid != current_process.pid:
+            os.kill(child.pid, 9)
+    os.kill(parent_process.pid, 9)
+
+
+def monkey_patch_vllm_p2p_access_check():
+    """
+    Monkey patch the slow p2p access check in vllm.
+    NOTE: We assume the p2p access is always allowed, which can be wrong for some setups.
+    """
+    import vllm.distributed.device_communicators.custom_all_reduce_utils as tgt
+
+    setattr(tgt, "gpu_p2p_access_check", lambda *arg, **kwargs: True)
+
+
 API_KEY_HEADER_NAME = "X-API-Key"
 
 
@@ -459,3 +481,4 @@ class APIKeyValidatorMiddleware(BaseHTTPMiddleware):
             )
         response = await call_next(request)
         return response
+