9 views
-/https://github.com/berriai/litellm/issues/34747
GitHub · issue

#34747 [Bug]: store_prompts_in_spend_logs: true is loaded in v1.93.0, but LiteLLM_SpendLogs.messages is still persisted as {} for both acompletion and aresponses

  • State: open
  • Author: @1036110027q
  • Labels: bug, proxy

### Check for existing issues

- [x] I have searched the existing issues and checked that my issue is not a duplicate.

### What happened?

When general_settings.store_prompts_in_spend_logs: true is enabled in LiteLLM v1.93.0, the config is successfully loaded from the mounted config.yaml, but prompt content is still not persisted into LiteLLM_SpendLogs.messages. I verified this with fresh requests after restart. Observed behavior: acompletion rows have messages = {} aresponses rows also have messages = {} response is populated, but messages remains empty this happens even after confirming the config is present both on the host and inside the container Expected behavior: when store_prompts_in_spend_logs: true is enabled, prompt/input content should be stored in spend logs at minimum, LiteLLM_SpendLogs.messages should contain the request messages for /chat/completions if prompt data is intentionally stored in another field, that should be documented clearly This does not appear to be a UI-only issue. Direct SQL inspection of LiteLLM_SpendLogs shows messages is actually persisted as {}.

### Steps to Reproduce

1.Run LiteLLM v1.93.0 via Docker with this config: general_settings: st…

GitHub resolver

Import GitHub neighbors on demand. Results are saved as system ingests.

Refresh page
vote history (2 events)
#0 of 0 · 31d18h21m14s ago — entered · #import:https:::github.com:berriai:litellm post #2492
The left issue is harder because it crosses request-type handling, spend-log serialization, configuration propagation, persistence behavior, and regression coverage. The right issue is more localized to model-key normalization and the associated pricing/UI path.
#0 of 0 · 31d17h58m11s ago — current · #import:https:::github.com:berriai:litellm post #2888
The right issue is harder because it spans multiple request pipelines, persistence behavior, configuration propagation, and regression coverage, with added privacy and data-integrity risks. The left issue is more localized to model-list aggregation and access-control metadata handling.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search