From 4a3819a6023e8ad4e0a29e19d52d47205d18223f Mon Sep 17 00:00:00 2001
From: ryan-crabbe-berri <ryan@berri.ai>
Date: Sat, 1 Aug 2026 16:55:54 -0700
Subject: [PATCH] fix(proxy_cli): import canonical proxy_server module in
 script-mode fallback

Running the proxy as a script (python litellm/proxy/proxy_cli.py) made the
relative-import fallback in run_server load proxy_server a second time as a
bare top-level module. That shadow copy won the litellm.callbacks
registration, while uvicorn served the canonical litellm.proxy.proxy_server
module, so budget/spend hooks incremented in-memory counters on one instance
and auth checks read them from another. Without Redis, key-level and
end-user model budgets were never enforced on such runs. Import the
canonical module in the fallback so only one module instance exists.
---
 litellm/proxy/proxy_cli.py                 |  3 +-
 tests/test_litellm/proxy/test_proxy_cli.py | 33 ++++++++++++++++++++++
 2 files changed, 34 insertions(+), 2 deletions(-)

diff --git a/litellm/proxy/proxy_cli.py b/litellm/proxy/proxy_cli.py
index cdfacae0a5d..ba9e70af5df 100644
--- a/litellm/proxy/proxy_cli.py
+++ b/litellm/proxy/proxy_cli.py
@@ -1013,8 +1013,7 @@ def run_server(
                 # user is missing a proxy dependency, ask them to pip install litellm[proxy]
                 raise e
             else:
-                # this is just a local/relative import error, user git cloned litellm
-                from proxy_server import (
+                from litellm.proxy.proxy_server import (
                     KeyManagementSettings,
                     ProxyConfig,
                     app,
diff --git a/tests/test_litellm/proxy/test_proxy_cli.py b/tests/test_litellm/proxy/test_proxy_cli.py
index 2ccaa0df440..789bf981274 100644
--- a/tests/test_litellm/proxy/test_proxy_cli.py
+++ b/tests/test_litellm/proxy/test_proxy_cli.py
@@ -2277,3 +2277,36 @@ class TestPostgresStatementTimeoutOptions:
                 standalone_mode=False,
             )
             return {k: os.environ[k] for k in ("DATABASE_URL", "DIRECT_URL") if k in os.environ}
+
+
+@pytest.mark.xdist_group("proxy_cli")
+class TestScriptModeImportsCanonicalProxyServer:
+    def test_no_shadow_proxy_server_module_in_script_mode(self):
+        """Script-mode runs (`python litellm/proxy/proxy_cli.py`) hit the relative-import
+        fallback in run_server. The old bare `from proxy_server import ...` loaded a second
+        full copy of proxy_server as a top-level module; that shadow copy won the
+        litellm.callbacks registration while uvicorn served the canonical module, so
+        in-memory budget counters were incremented on one instance and read on another
+        and budgets were never enforced without Redis."""
+        import subprocess
+
+        repo_root = Path(__file__).resolve().parents[3]
+        driver = (
+            "import sys, runpy\n"
+            "sys.path.insert(0, 'litellm/proxy')\n"
+            "sys.argv = ['proxy_cli.py', '--version']\n"
+            "try:\n"
+            "    runpy.run_path('litellm/proxy/proxy_cli.py', run_name='__main__')\n"
+            "except SystemExit:\n"
+            "    pass\n"
+            "assert 'litellm.proxy.proxy_server' in sys.modules, 'canonical module not loaded'\n"
+            "assert 'proxy_server' not in sys.modules, 'shadow top-level proxy_server module loaded'\n"
+        )
+        result = subprocess.run(
+            [sys.executable, "-c", driver],
+            cwd=repo_root,
+            capture_output=True,
+            text=True,
+            timeout=120,
+        )
+        assert result.returncode == 0, f"stdout={result.stdout}\nstderr={result.stderr}"