fix(proxy): return early instead of raising ValueError when standard_logging_payload is missing

[themavik]

Summary

Fixes #18986

The _PROXY_VirtualKeyModelMaxBudgetLimiter.async_log_success_event callback raises a ValueError when standard_logging_payload is None. This happens for non-standard endpoints like the vLLM /classify passthrough, which do not populate standard_logging_object in kwargs.

The raised exception:

1. Produces noisy error logs (LiteLLM.LoggingError: [Non-Blocking])
2. Disrupts downstream success callbacks (e.g. Langfuse fails to log the request)

Changes

Changed the raise ValueError to an early return with a debug log, matching the existing defensive pattern used a few lines later for missing user_api_key_model_max_budget (lines 163-171 of the same method).

Test plan

• vLLM /classify endpoint calls should no longer produce ValueError: standard_logging_payload is required in logs
• Downstream callbacks like Langfuse should run normally for these endpoints
• Standard chat/completion endpoints are unaffected since they always populate standard_logging_payload

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
litellm	Ready	Preview, Comment	Feb 10, 2026 10:41am

[greptile-apps]

Greptile Overview

Greptile Summary

This PR updates the proxy hook _PROXY_VirtualKeyModelMaxBudgetLimiter.async_log_success_event to treat a missing standard_logging_payload as a non-actionable condition: it now logs at debug level and returns early rather than raising ValueError. This aligns with the method’s existing defensive behavior for other missing optional inputs (e.g., absent max-budget metadata) and prevents the hook from emitting noisy non-blocking logging errors.

In the broader proxy callback chain, this change ensures non-standard/passthrough endpoints that don’t populate the standard logging object (e.g., vLLM /classify) won’t interrupt the remaining success callbacks (such as Langfuse logging), while standard chat/completion flows continue unchanged because they already provide standard_logging_payload.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk.
• The change is narrowly scoped (error-on-missing-input to early return) within a non-blocking logging hook, aligns with existing defensive patterns in the same method, and reduces unwanted exceptions without affecting standard request flows that already provide the payload.
• No files require special attention

Important Files Changed

Filename	Overview
litellm/proxy/hooks/model_max_budget_limiter.py	Replaces a ValueError on missing standard_logging_payload with an early return and debug log in async_log_success_event to avoid noisy errors and allow downstream callbacks to proceed.

Sequence Diagram

sequenceDiagram
    participant Proxy as LiteLLM Proxy
    participant Hook as _PROXY_VirtualKeyModelMaxBudgetLimiter
    participant Other as Downstream Success Callbacks

    Proxy->>Hook: async_log_success_event(kwargs)
    alt standard_logging_payload missing (None)
        Hook->>Hook: debug log + early return
        Hook-->>Proxy: returns (no exception)
        Proxy->>Other: continue running callbacks
    else standard_logging_payload present
        Hook->>Hook: extract user_api_key_model_max_budget
        alt max budget missing
            Hook->>Hook: debug log + early return
        else max budget present
            Hook->>Hook: update usage/budget state
        end
        Hook-->>Proxy: returns
        Proxy->>Other: continue running callbacks
    end

Loading

[greptile-apps]

_{1 file reviewed, no comments}

_{Edit Code Review Agent Settings | Greptile}

[krrishdholakia]

Hi @themavik thank you for your work. Are you looking for new opportunities at the moment?

[themavik]

Hi @themavik thank you for your work. Are you looking for new opportunities at the moment?

Hi @krrishdholakia , thank you for reaching out! Yes, I am currently open to new opportunities and would love to hear more about what you have in mind.