[GitHub] [tvm] zxybazh commented on a diff in pull request #10874: [BugFix] Fix PopenWorker Memory Leak

[GitBox <gi...@apache.org>]

zxybazh commented on code in PR #10874:
URL: https://github.com/apache/tvm/pull/10874#discussion_r840925066


##########
python/tvm/exec/popen_worker.py:
##########
@@ -40,22 +40,34 @@ def main():
     if len(sys.argv) != 3:
         print("Usage: <read_fd> <write_fd>")
         return
-    if sys.platform == "win32":
-        # pylint: disable=import-outside-toplevel
-        import msvcrt
-
-        reader = os.fdopen(msvcrt.open_osfhandle(int(sys.argv[1]), os.O_BINARY), "rb")
-        writer = os.fdopen(msvcrt.open_osfhandle(int(sys.argv[2]), os.O_BINARY), "wb")
-    else:
-        reader = os.fdopen(int(sys.argv[1]), "rb")
-        writer = os.fdopen(int(sys.argv[2]), "wb")
 
     logging.basicConfig(level=logging.INFO)
 
     lock = threading.Lock()
 
+    def get_reader():

Review Comment:
   Hi TQ, it does not acutally help that we put `fdopen` into a subfunction. <s>The cause of the memory leak is acutally reusing the same `reader` and `writer`. The change is acutally to create new `reader` and `writer` everytime instead of reusing the same ones to avoid leak.</s> After checking the proccess status, I found that this fix does not really fix the problem because it only caused the process to suicide which avoided memory leak.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@tvm.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org