From b6cf4066f4e907c03f11065f52f4da149e649128 Mon Sep 17 00:00:00 2001
From: shivam <shivam@berri.ai>
Date: Sat, 25 Jul 2026 01:17:43 +0000
Subject: [PATCH] fix(caching): log cached anthropic stream replay only once

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---
 .../messages/response_cache.py                |  5 ++-
 .../messages/test_response_cache.py           | 32 +++++++++++++++++++
 2 files changed, 36 insertions(+), 1 deletion(-)

diff --git a/litellm/llms/anthropic/experimental_pass_through/messages/response_cache.py b/litellm/llms/anthropic/experimental_pass_through/messages/response_cache.py
index 4873b1fdb96..d5b68c99130 100644
--- a/litellm/llms/anthropic/experimental_pass_through/messages/response_cache.py
+++ b/litellm/llms/anthropic/experimental_pass_through/messages/response_cache.py
@@ -96,6 +96,7 @@ class CachedAnthropicMessagesStreamIterator(BaseAnthropicMessagesStreamingIterat
         super().__init__(litellm_logging_obj=litellm_logging_obj, request_body=request_body)
         self.chunks: list[bytes] = [event.encode("utf-8") for event in events]
         self.current_index = 0
+        self.logged = False
         self._hidden_params: dict[str, Any] = {"cache_hit": True}
         litellm_logging_obj.model_call_details["cache_hit"] = True
 
@@ -104,7 +105,9 @@ class CachedAnthropicMessagesStreamIterator(BaseAnthropicMessagesStreamingIterat
 
     async def __anext__(self) -> bytes:
         if self.current_index >= len(self.chunks):
-            await self._handle_streaming_logging(self.chunks)
+            if not self.logged:
+                self.logged = True
+                await self._handle_streaming_logging(self.chunks)
             raise StopAsyncIteration
         chunk = self.chunks[self.current_index]
         self.current_index += 1
diff --git a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_response_cache.py b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_response_cache.py
index 344152cd828..071580347a6 100644
--- a/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_response_cache.py
+++ b/tests/test_litellm/llms/anthropic/experimental_pass_through/messages/test_response_cache.py
@@ -177,3 +177,35 @@ async def test_abandoned_stream_is_not_cached(local_cache, request_kwargs, monke
 
     assert len(fake_handler.calls) == 2
     assert replayed == STREAM_EVENTS
+
+@pytest.mark.asyncio
+async def test_cached_stream_replay_logs_once_when_polled_after_exhaustion():
+    from unittest.mock import AsyncMock, MagicMock, patch
+
+    from litellm.llms.anthropic.experimental_pass_through.messages.response_cache import (
+        CachedAnthropicMessagesStreamIterator,
+    )
+    from litellm.proxy.pass_through_endpoints.streaming_handler import (
+        PassThroughStreamingHandler,
+    )
+
+    logging_obj = MagicMock()
+    logging_obj.model_call_details = {}
+    iterator = CachedAnthropicMessagesStreamIterator(
+        events=[event.decode("utf-8") for event in STREAM_EVENTS],
+        litellm_logging_obj=logging_obj,
+        request_body={"model": "claude-sonnet-4-5"},
+    )
+
+    with patch.object(
+        PassThroughStreamingHandler,
+        "_route_streaming_logging_to_handler",
+        new=AsyncMock(),
+    ) as mock_route:
+        assert await _collect(iterator) == STREAM_EVENTS
+        for _ in range(2):
+            with pytest.raises(StopAsyncIteration):
+                await iterator.__anext__()
+        await asyncio.sleep(0)
+
+    mock_route.assert_called_once()