16 views
-/https://github.com/berriai/litellm/issues/15647
GitHub · issue

#15647 [Bug]: Updated fallback configuration not reloaded on redeploy

  • State: open
  • Author: @ppiotrlach
  • Labels: bug

### What happened?

Bug: Updated fallback configuration not reloaded on redeploy

## Summary When redeploying an existing LiteLLM setup with updated configuration, changes in the fallbacks section are not applied unless manually updated via the LiteLLM API.

## Details After adding a new model (claude-4-5-sonnet) along with its fallback configuration, the changes were correctly applied in our validation environment (which is deployed from scratch). However, in development and production environments, where the infrastructure already existed, the updated fallback configuration was not reflected after redeployment. The previous fallback mappings remained active until I manually reconfigured them in the UI.

## Expected behavior LiteLLM should detect and apply changes to the fallback configuration (fallbacks) on redeploy, even when the infrastructure already exist (database).

## Actual behavior Redeploying LiteLLM with an updated config does not refresh the fallback section for existing environments.

## Reproduction steps

- Deploy LiteLLM with an initial fallback configuration. - Update the config file to include a new model and fallback mapping. - Redeploy the application (without…

GitHub resolver

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

Refresh page
vote history (3 events)
#0 of 0 · 31d18h43m19s ago — entered · #import:https:::github.com:berriai:litellm post #2164
The left issue spans configuration reconciliation, redeploy lifecycle behavior, persistence, and backward-compatibility concerns, requiring broader integration testing. The right issue is comparatively localized to callback metadata representation and assignment logic.
#29914 is harder because it spans multiple API surfaces, provider integration, OCR-specific accounting, and compatibility/testing work. #15647 is narrower, mainly requiring diagnosis and correction of configuration synchronization during redeployment.
#0 of 0 · 31d18h30m1s ago — current · #import:https:::github.com:berriai:litellm post #2385
The right issue is harder because it requires reliable configuration reconciliation across redeploys, persistence, startup lifecycle, and backward-compatibility concerns, while the left issue is primarily a UI data-filtering correction against an already-enforced API behavior.
discussed in #import:https:::github.com:berriai:litellm

ranked child groups

no voted pairs yet in this scope

cli
src
spread
search