From 8cb42ca7b4a61bf6d32f4b3a0a21ab557e84fceb Mon Sep 17 00:00:00 2001
From: wzj1228516103 <154067345+wzj1228516103@users.noreply.github.com>
Date: Tue, 8 Sep 2026 18:21:56 +0800
Subject: [PATCH] fix(proxy): increase spend counter cache capacity

---
 litellm/proxy/proxy_server.py                 |  7 ++++++-
 tests/test_litellm/proxy/test_proxy_server.py | 18 ++++++++++++++++++
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 32b6b841af7a..69b3b890c226 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -240,6 +240,7 @@ def generate_feedback_box():
 from litellm import Router
 from litellm._logging import _redact_string, verbose_proxy_logger, verbose_router_logger
 from litellm.caching.caching import DualCache, RedisCache
+from litellm.caching.in_memory_cache import InMemoryCache
 from litellm.caching.redis_cluster_cache import RedisClusterCache
 from litellm.constants import (
     _REALTIME_BODY_CACHE_SIZE,
@@ -2301,7 +2302,11 @@ async def root_redirect():
 user_api_key_cache: UserApiKeyCache = UserApiKeyCache(
     default_in_memory_ttl=UserAPIKeyCacheTTLEnum.in_memory_cache_ttl.value
 )
-spend_counter_cache: Final = DualCache(default_in_memory_ttl=UserAPIKeyCacheTTLEnum.in_memory_cache_ttl.value)
+SPEND_COUNTER_CACHE_MAX_SIZE: Final = 10_000
+spend_counter_cache: Final = DualCache(
+    in_memory_cache=InMemoryCache(max_size_in_memory=SPEND_COUNTER_CACHE_MAX_SIZE),
+    default_in_memory_ttl=UserAPIKeyCacheTTLEnum.in_memory_cache_ttl.value,
+)
 cli_sso_session_cache: Final = DualCache(default_in_memory_ttl=CLI_SSO_SESSION_TTL_SECONDS)
 model_max_budget_limiter: Final = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=spend_counter_cache)
 litellm.logging_callback_manager.add_litellm_callback(model_max_budget_limiter)
diff --git a/tests/test_litellm/proxy/test_proxy_server.py b/tests/test_litellm/proxy/test_proxy_server.py
index b7bb58378d40..63ce73b3224a 100644
--- a/tests/test_litellm/proxy/test_proxy_server.py
+++ b/tests/test_litellm/proxy/test_proxy_server.py
@@ -7542,6 +7542,24 @@ async def test_store_model_in_db_db_failure_graceful(monkeypatch):
 # =====================================================================
 
 
+def test_spend_counter_cache_keeps_active_counters_warm(monkeypatch):
+    from litellm.caching.in_memory_cache import InMemoryCache
+    from litellm.proxy.proxy_server import spend_counter_cache
+
+    cache = InMemoryCache(
+        max_size_in_memory=spend_counter_cache.in_memory_cache.max_size_in_memory,
+        default_ttl=60,
+    )
+    monkeypatch.setattr(spend_counter_cache, "in_memory_cache", cache)
+
+    sentinel = "spend:key:sentinel"
+    cache.set_cache(key=sentinel, value=1.0)
+    for index in range(300):
+        cache.set_cache(key=f"spend:end_user:scope-{index}", value=1.0)
+
+    assert cache.get_cache(key=sentinel) == 1.0
+
+
 @pytest.mark.asyncio
 async def test_get_current_spend_reads_redis_first():
     """get_current_spend should prefer Redis over in-memory."""