From 1f7a4806ff41d270f94e0fa1f58ca09a70a087e3 Mon Sep 17 00:00:00 2001
From: Simon Mo <simon.mo@hey.com>
Date: Thu, 3 Dec 2020 08:45:04 -0800
Subject: [PATCH] [Serve] Fix Flask Request self reference (#12560)

* [Serve] Fix Flask Request self reference

* Working flag

* Fix
---
 python/ray/serve/http_util.py             |  5 +++-
 python/ray/serve/tests/test_regression.py | 28 +++++++++++++++++++++++
 2 files changed, 32 insertions(+), 1 deletion(-)

diff --git a/python/ray/serve/http_util.py b/python/ray/serve/http_util.py
index 177a89f4b..da41c10eb 100644
--- a/python/ray/serve/http_util.py
+++ b/python/ray/serve/http_util.py
@@ -11,7 +11,10 @@ def build_flask_request(asgi_scope_dict, request_body):
     happen.
     """
     wsgi_environ = build_wsgi_environ(asgi_scope_dict, request_body)
-    return flask.Request(wsgi_environ)
+    # We set populate_request=False to prevent self reference, which can lead
+    # to objects tracked by python garbage collector and memory growth. See
+    # https://github.com/ray-project/ray/issues/12395.
+    return flask.Request(wsgi_environ, populate_request=False)
 
 
 def build_wsgi_environ(scope, body):
diff --git a/python/ray/serve/tests/test_regression.py b/python/ray/serve/tests/test_regression.py
index 010478211..027b53a72 100644
--- a/python/ray/serve/tests/test_regression.py
+++ b/python/ray/serve/tests/test_regression.py
@@ -1,6 +1,9 @@
+import gc
+
 import numpy as np
 import requests
 
+import ray
 from ray import serve
 
 
@@ -35,6 +38,31 @@ def test_np_in_composed_model(serve_instance):
     assert result.json() == 100.0
 
 
+def test_backend_worker_memory_growth(serve_instance):
+    # https://github.com/ray-project/ray/issues/12395
+    client = serve_instance
+
+    def gc_unreachable_objects(flask_request):
+        gc.set_debug(gc.DEBUG_SAVEALL)
+        gc.collect()
+        return len(gc.garbage)
+
+    client.create_backend("model", gc_unreachable_objects)
+    client.create_endpoint("model", backend="model", route="/model")
+
+    handle = client.get_handle("model")
+
+    for _ in range(10):
+        result = requests.get("http://127.0.0.1:8000/model")
+        assert result.status_code == 200
+        num_unreachable_objects = result.json()
+        assert num_unreachable_objects == 0
+
+    for _ in range(10):
+        num_unreachable_objects = ray.get(handle.remote())
+        assert num_unreachable_objects == 0
+
+
 if __name__ == "__main__":
     import sys
     import pytest