Skip to content

enterprise hook: generic 4xx still returns type/param as the string "None"; allow vLLM 'max_tokens must be at most N' through #121

Closed

Description

@sre-helmcode

Follow-ups from the nan-devops#636 reviews:

  1. Enterprise type/param bug. The enterprise usage hook still returns the generic sanitised 4xx with "type":"None","param":"None", the literal string. Community fixed this in #636: param is JSON null and type is mapped by status. Port the same mapping to the enterprise hook, which is a separate program, and tighten its parity test afterwards.
  2. 5xx type. In both tenants 5xx still carries "type":"None"; map it to server_error.
  3. 429 type consistency. Our own 429s now say rate_limit_error, while upstream-origin ones keep throttling_error. Pick one.
  4. Useful vLLM message is swallowed. The generic sanitiser replaces vLLM's max_tokens must be at most N (no vendor information in it) with the generic 400. Let that message through.

🤖 Generated with Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions