From 02fd30121a995552dd857ee1b81bf602b112a94a Mon Sep 17 00:00:00 2001
From: Prasanna Jain <prassu017@gmail.com>
Date: Wed, 28 Oct 2026 01:25:01 -0400
Subject: [PATCH] fix(caching): keep parent_otel_span out of the response cache
 key

---
 litellm/types/litellm_params.py         |  2 ++
 tests/unit/caching/test_caching.py      | 10 ++++++++++
 tests/unit/types/test_litellm_params.py |  3 +++
 3 files changed, 15 insertions(+)

diff --git a/litellm/types/litellm_params.py b/litellm/types/litellm_params.py
index 717b74e5741..88130646318 100644
--- a/litellm/types/litellm_params.py
+++ b/litellm/types/litellm_params.py
@@ -14,6 +14,7 @@ if TYPE_CHECKING:
     import httpx
     from aiohttp import ClientSession
     from openai import AsyncAzureOpenAI, AsyncOpenAI, AzureOpenAI, OpenAI
+    from opentelemetry.trace import Span
 
     from litellm.litellm_core_utils.litellm_logging import Logging
     from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, HTTPHandler
@@ -341,6 +342,7 @@ class CallState:
     model_alias_map: Mapping[str, str] | None = None
     data_residency: str | None = None
     litellm_logging_obj: "Logging | None" = None
+    parent_otel_span: "Span | None" = None
     preset_cache_key: str | None = None
     cache_key: str | None = None
     stream_response: "Mapping[str, ModelResponse] | None" = None
diff --git a/tests/unit/caching/test_caching.py b/tests/unit/caching/test_caching.py
index 91083921130..418e75deb08 100644
--- a/tests/unit/caching/test_caching.py
+++ b/tests/unit/caching/test_caching.py
@@ -445,6 +445,16 @@ def test_provider_specific_cache_key_ignores_litellm_owned_kwargs(monkeypatch: p
     assert cache.get_cache_key(**{**request, "top_k": 6}) != base_key
 
 
+def test_provider_specific_cache_key_ignores_parent_otel_span(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setattr(litellm, "enable_caching_on_provider_specific_optional_params", True)
+    cache: Final = Cache(type=LiteLLMCacheType.LOCAL)
+    request: Final = {"model": "gpt-4.1-mini", "messages": [{"role": "user", "content": "hi"}]}
+
+    assert cache.get_cache_key(**request, parent_otel_span=object()) == cache.get_cache_key(
+        **request, parent_otel_span=object()
+    )
+
+
 def test_cache_key_tells_apart_requests_that_differ_only_in_continuation_token() -> None:
     cache: Final = Cache(type=LiteLLMCacheType.LOCAL)
     request: Final = {
diff --git a/tests/unit/types/test_litellm_params.py b/tests/unit/types/test_litellm_params.py
index 740342f54d9..c0122ae9057 100644
--- a/tests/unit/types/test_litellm_params.py
+++ b/tests/unit/types/test_litellm_params.py
@@ -9,6 +9,7 @@ import httpx
 import pytest
 from aiohttp import ClientSession
 from openai import AsyncAzureOpenAI, AsyncOpenAI, AzureOpenAI, OpenAI
+from opentelemetry.trace import Span
 from pydantic import BaseModel, ConfigDict, TypeAdapter, ValidationError
 
 import litellm
@@ -249,6 +250,7 @@ INTERNAL_STATE_NAMES: Final = (
     "model_alias_map",
     "data_residency",
     "litellm_logging_obj",
+    "parent_otel_span",
     "preset_cache_key",
     "cache_key",
     "stream_response",
@@ -524,6 +526,7 @@ TYPE_HINT_NAMESPACE: Final[Mapping[str, object]] = {
     "ModelResponse": ModelResponse,
     "ModelResponseStream": ModelResponseStream,
     "Logging": Logging,
+    "Span": Span,
     "SecretFields": SecretFields,
     "CompactionState": CompactionState,
     "RouterWeights": RouterWeights,