Add Prometheus metric to track callback logging failures in S3

[Sameerlite]

Title

Add Prometheus metric to track callback logging failures

Relevant issues

Adds monitoring for callback health - tracks when S3, Langfuse, and other callbacks fail to log events.

Pre-Submission checklist

• I have Added testing in the tests/litellm/ directory
• I have added a screenshot of my new test passing locally
• My PR passes all unit tests on make test-unit
• My PR's scope is as isolated as possible, it only solves 1 specific problem

Type

🆕 New Feature

Changes

New Prometheus Metric

Added litellm_callback_logging_failures_metric to track when callbacks (S3, Langfuse, etc.) fail to log events.

Metric:

• Name: litellm_callback_logging_failures_metric
• Type: Counter
• Label: callback_name (e.g., "S3Logger", "LangFuseLogger")

Example:

litellm_callback_logging_failures_metric_total{callback_name="S3Logger"} 5.0
litellm_callback_logging_failures_metric_total{callback_name="LangFuseLogger"} 2.0

Files Modified

1. enterprise/litellm_enterprise/integrations/prometheus.py

   • Added metric definition (line 302-306)
   • Added increment_callback_logging_failure() method (line 1733-1750)

2. litellm/integrations/custom_logger.py

   • Added handle_callback_failure() method that all callbacks can use (line 571-624)

3. litellm/integrations/s3_v2.py

   • Modified exception handlers to call handle_callback_failure() on upload failures
   • Tracks failures in async and sync upload methods
   • Changes here were done because we don't raise the error from here. So it makes sense to just call the method here itself

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Preview	Comments	Updated (UTC)
litellm	Error			Oct 31, 2025 11:40am

[krrishdholakia]

which would be less work over time:

• requiring each instance to implement this
• OR having integrations just bubble the error and have litellm_logging handle this?
  @Sameerlite

[Sameerlite]

@krrishdholakia 2nd one is less work but the problem is periodic_flush and all method used in it don't raise error or propagate it to litellm_logging. Plus there are tasks which are fire-and-forget which I wasn't able find a way to bubble up those errors. The method I used was making sure that if an error comes, it will get logged in Prometheus