From 48b95aa0c733c2ecc97624851c50ce94bb292a9d Mon Sep 17 00:00:00 2001
From: xuan <118178092+yuee3@users.noreply.github.com>
Date: Mon, 12 Oct 2026 13:45:04 +0800
Subject: [PATCH] fix(prometheus): give each proxy boot its own auto-created
 PROMETHEUS_MULTIPROC_DIR

---
 litellm/proxy/proxy_cli.py                  | 14 +++---
 tests/unit/proxy/test_prometheus_cleanup.py | 47 +++++++++++++++++++++
 2 files changed, 56 insertions(+), 5 deletions(-)

diff --git a/litellm/proxy/proxy_cli.py b/litellm/proxy/proxy_cli.py
index c7371f294f0..2e677857573 100644
--- a/litellm/proxy/proxy_cli.py
+++ b/litellm/proxy/proxy_cli.py
@@ -724,12 +724,16 @@ class ProxyInitializationHelpers:
                 wipe_directory(configured_dir)
             return None
 
-        multiproc_dir: Final = configured_dir or os.path.join(tempfile.gettempdir(), "litellm_prometheus_multiproc")
-        os.environ["PROMETHEUS_MULTIPROC_DIR"] = multiproc_dir
+        if configured_dir:
+            # An operator-set directory is theirs to manage; keep the historical wipe-on-boot.
+            os.makedirs(configured_dir, exist_ok=True)
+            wipe_directory(configured_dir)
+            multiproc_dir, action = configured_dir, "Using existing"
+        else:
+            # Unique per boot: a fixed shared name lets a second proxy on this host wipe the first one's counters.
+            multiproc_dir, action = tempfile.mkdtemp(prefix="litellm_prometheus_multiproc_"), "Auto-created"
 
-        os.makedirs(multiproc_dir, exist_ok=True)
-        wipe_directory(multiproc_dir)
-        action: Final = "Using existing" if configured_dir else "Auto-created"
+        os.environ["PROMETHEUS_MULTIPROC_DIR"] = multiproc_dir
         print(f"LiteLLM: {action} PROMETHEUS_MULTIPROC_DIR={multiproc_dir}")
         return multiproc_dir
 
diff --git a/tests/unit/proxy/test_prometheus_cleanup.py b/tests/unit/proxy/test_prometheus_cleanup.py
index c774d376d33..95903540c47 100644
--- a/tests/unit/proxy/test_prometheus_cleanup.py
+++ b/tests/unit/proxy/test_prometheus_cleanup.py
@@ -6,8 +6,10 @@ ProxyInitializationHelpers._maybe_setup_prometheus_multiproc_dir.
 from __future__ import annotations
 
 import os
+import shutil
 import subprocess
 import sys
+import tempfile
 from pathlib import Path
 from typing import Final
 from unittest.mock import patch
@@ -193,6 +195,8 @@ class TestMaybeSetupPrometheusMultiprocDir:
             result_dir = os.environ.get("PROMETHEUS_MULTIPROC_DIR")
             assert result_dir is not None
             assert os.path.isdir(result_dir)
+            legacy_shared = os.path.join(tempfile.gettempdir(), "litellm_prometheus_multiproc")
+            assert result_dir != legacy_shared
 
             # Cleanup
             os.environ.pop("PROMETHEUS_MULTIPROC_DIR", None)
@@ -257,3 +261,46 @@ class TestMaybeSetupPrometheusMultiprocDir:
             assert os.environ["PROMETHEUS_MULTIPROC_DIR"] == str(tmp_path)
         assert not admitted.exists()
         assert not samples.exists()
+
+    def test_auto_created_dir_is_unique_per_boot_and_keeps_sibling_counters(self):
+        """A fixed default dir made the second proxy boot wipe the first proxy's live counters (issue #42566)."""
+        with patch.dict(os.environ, {}, clear=False):
+            os.environ.pop("PROMETHEUS_MULTIPROC_DIR", None)
+            os.environ.pop("prometheus_multiproc_dir", None)
+            litellm_settings = {"callbacks": ["prometheus"]}
+
+            first = ProxyInitializationHelpers._maybe_setup_prometheus_multiproc_dir(
+                num_workers=4, litellm_settings=litellm_settings
+            )
+            assert first is not None and os.path.isdir(first)
+
+            stale = os.path.join(first, "counter_proxy_one.db")
+            with open(stale, "w") as f:
+                f.write("live")
+
+            os.environ.pop("PROMETHEUS_MULTIPROC_DIR", None)  # simulate a second, independent boot
+            second = ProxyInitializationHelpers._maybe_setup_prometheus_multiproc_dir(
+                num_workers=4, litellm_settings=litellm_settings
+            )
+            assert second is not None and os.path.isdir(second)
+
+            assert first != second
+            assert os.path.exists(stale)
+
+            os.environ.pop("PROMETHEUS_MULTIPROC_DIR", None)
+            shutil.rmtree(first, ignore_errors=True)
+            shutil.rmtree(second, ignore_errors=True)
+
+    def test_configured_dir_is_still_wiped_on_boot(self, tmp_path):
+        """A directory the operator set keeps its historical wipe-on-boot lifecycle."""
+        stale = tmp_path / "counter_stale.db"
+        stale.write_text("old")
+
+        with patch.dict(os.environ, {"PROMETHEUS_MULTIPROC_DIR": str(tmp_path)}):
+            result_dir = ProxyInitializationHelpers._maybe_setup_prometheus_multiproc_dir(
+                num_workers=4,
+                litellm_settings={"callbacks": ["prometheus"]},
+            )
+
+            assert result_dir == str(tmp_path)
+            assert not stale.exists()