[Bug]: turn_off_message_logging sometimes not redacting output

[clarity99]

What happened?

I am running litellm proxy in docker and have a langfuse integration turned on and I notice that sometimes, the messages are not redacted there, specifically the AI output (user input seems to always be redacted).
This only happens sometimes and I have no idea what triggers it or how to debug it.

settings:
litellm_settings:
modify_params: true
cache: false
success_callback: ["langfuse"]
failure_callback: ["sentry"]
turn_off_message_logging: True

json output in langfuse:

Input

{
messages: [
0: {
role: "user"
content: "redacted-by-litellm"
}
]
}
Output

{
content: "Seveda, z veseljem ti pomagam. Tukaj je nekaj možnosti, kako bi lahko jedrnato povzel tvoje delo s klientko:

Relevant log output

20:43:56 - LiteLLM Proxy:INFO: parallel_request_limiter.py:68 - Current Usage of key in this minute: None
litellm-1  | 20:43:56 - LiteLLM Proxy:INFO: parallel_request_limiter.py:68 - Current Usage of user in this minute: None
litellm-1  | 20:43:56 - LiteLLM:INFO: utils.py:3035 - 
litellm-1  | LiteLLM completion() model= gemini-2.0-pro-exp-02-05; provider = vertex_ai
litellm-1  | 20:43:56 - LiteLLM:INFO: cost_calculator.py:588 - selected model name for cost calculation: vertex_ai/gemini-2.0-pro-exp-02-05
litellm-1  | 20:43:56 - LiteLLM Router:INFO: router.py:1051 - litellm.acompletion(model=vertex_ai/gemini-2.0-pro-exp-02-05) 200 OK
litellm-1  | INFO:     172.18.0.6:38458 - "POST /v1/chat/completions HTTP/1.1" 200 OK
litellm-1  | 20:44:05 - LiteLLM:INFO: cost_calculator.py:588 - selected model name for cost calculation: vertex_ai/gemini-2.0-pro-exp-02-05
litellm-1  | 20:44:05 - LiteLLM Proxy:INFO: proxy_server.py:899 - Writing spend log to db - request_id: chatcmpl-a98d9f79-38be-43a4-94d7-206b253839d1, spend: 0.0
litellm-1  | 20:44:05 - LiteLLM:INFO: cost_calculator.py:588 - selected model name for cost calculation: vertex_ai/gemini-2.0-pro-exp-02-05
litellm-1  | 20:44:05 - LiteLLM:INFO: langfuse.py:261 - Langfuse Layer Logging - logging success

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

1.63.14

Twitter / LinkedIn details

No response