diff --git a/litellm/caching/redis_cache.py b/litellm/caching/redis_cache.py
index f1c80eaacbe..90adda6b1cc 100644
--- a/litellm/caching/redis_cache.py
+++ b/litellm/caching/redis_cache.py
@@ -1370,7 +1370,8 @@ class RedisCache(BaseCache):
         self.redis_client.flushall()
 
     async def disconnect(self):
-        await self.async_redis_conn_pool.disconnect(inuse_connections=True)
+        if self.async_redis_conn_pool is not None:
+            await self.async_redis_conn_pool.disconnect(inuse_connections=True)
         try:
             self.redis_client.close()
         except Exception as e:
diff --git a/litellm/caching/redis_cluster_cache.py b/litellm/caching/redis_cluster_cache.py
index 12d285ca5a8..52a3a5696de 100644
--- a/litellm/caching/redis_cluster_cache.py
+++ b/litellm/caching/redis_cluster_cache.py
@@ -56,6 +56,11 @@ class RedisClusterCache(RedisCache):
         async_redis_cluster_client: Final = self.init_async_client()
         return await async_redis_cluster_client.mget_nonatomic(keys=keys)
 
+    async def disconnect(self):
+        if self.redis_async_redis_cluster_client is not None:
+            await self.redis_async_redis_cluster_client.aclose()
+        await super().disconnect()
+
     async def test_connection(self) -> dict:
         """
         Test the Redis Cluster connection.
diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 990682f10a5..cac8a9d6893 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -908,7 +908,10 @@ async def proxy_shutdown_event(worker_heartbeat: ProxyWorkerHeartbeat | None = N
         await prisma_client.disconnect()
 
     if litellm.cache is not None:
-        await litellm.cache.disconnect()
+        try:
+            await litellm.cache.disconnect()
+        except Exception as e:
+            verbose_proxy_logger.debug("Error disconnecting litellm.cache: %s", e)
 
     await jwt_handler.close()
 
diff --git a/tests/test_litellm/caching/test_redis_cache.py b/tests/test_litellm/caching/test_redis_cache.py
index 487a64797d1..16f527a5350 100644
--- a/tests/test_litellm/caching/test_redis_cache.py
+++ b/tests/test_litellm/caching/test_redis_cache.py
@@ -465,6 +465,21 @@ def test_delete_cache_namespaces_key(namespace, expected, monkeypatch, redis_no_
     mock_client.delete.assert_called_once_with(expected)
 
 
+@pytest.mark.asyncio
+async def test_disconnect_with_no_connection_pool_does_not_raise():
+    """RedisCache.disconnect() must not blow up when async_redis_conn_pool is None,
+    which is the case in cluster mode (get_redis_connection_pool returns None for
+    startup_nodes configs). Regression test for
+    https://github.com/BerriAI/litellm/issues/37137."""
+    redis_cache = RedisCache.__new__(RedisCache)
+    redis_cache.async_redis_conn_pool = None
+    redis_cache.redis_client = MagicMock()
+
+    await redis_cache.disconnect()
+
+    redis_cache.redis_client.close.assert_called_once()
+
+
 def _closed_port() -> int:
     """A port with nothing listening, so Redis calls fail fast and deterministically."""
     import socket
diff --git a/tests/test_litellm/caching/test_redis_cluster_cache.py b/tests/test_litellm/caching/test_redis_cluster_cache.py
index 372425aa9fa..01ee4a9604b 100644
--- a/tests/test_litellm/caching/test_redis_cluster_cache.py
+++ b/tests/test_litellm/caching/test_redis_cluster_cache.py
@@ -1,5 +1,7 @@
 import json
-from unittest.mock import MagicMock, patch
+import os
+import sys
+from unittest.mock import AsyncMock, MagicMock, patch
 
 import pytest
 from fastapi.testclient import TestClient
@@ -116,6 +118,44 @@ def test_cache_init_creates_redis_cache_without_cluster_config(
     assert not isinstance(cache.cache, RedisClusterCache)
 
 
+@pytest.mark.asyncio
+@patch("litellm._redis.init_redis_cluster")
+async def test_disconnect_closes_cluster_client_without_raising(mock_init_redis_cluster):
+    """Shutting down a proxy in cluster mode used to always raise AttributeError,
+    since RedisCache.disconnect() unconditionally dereferences async_redis_conn_pool,
+    which is None by design in cluster mode. Regression test for
+    https://github.com/BerriAI/litellm/issues/37137."""
+    cache = RedisClusterCache(
+        startup_nodes=[{"host": "localhost", "port": 6379}],
+        password="hello",
+    )
+    assert cache.async_redis_conn_pool is None
+
+    mock_cluster_client = AsyncMock()
+    cache.redis_async_redis_cluster_client = mock_cluster_client
+    cache.redis_client = MagicMock()
+
+    await cache.disconnect()
+
+    mock_cluster_client.aclose.assert_awaited_once()
+
+
+@pytest.mark.asyncio
+@patch("litellm._redis.init_redis_cluster")
+async def test_disconnect_without_cluster_client_does_not_raise(mock_init_redis_cluster):
+    """disconnect() must be a no-op for the cluster client when it was never
+    initialized (e.g. shutdown before any request touched Redis)."""
+    cache = RedisClusterCache(
+        startup_nodes=[{"host": "localhost", "port": 6379}],
+        password="hello",
+    )
+    cache.redis_client = MagicMock()
+
+    await cache.disconnect()
+
+    cache.redis_client.close.assert_called_once()
+
+
 @pytest.mark.parametrize(
     "startup_nodes, env_var, expected_cache_type",
     [