#  Copyright (c) 2026 Cisco Systems, Inc. and its affiliates
#  SPDX-License-Identifier: Apache-2.0
$schema: "https://json-schema.org/draft-07/schema#"
$id: "infra/v1#llm-scheduler-params"
title: LlmSchedulerMiddlewareParams
description: >
  Constructor kwargs for the ``llm_scheduler`` infra middleware plugin.
  Selected when the sibling ``middleware`` id is ``llm_scheduler``
  (aliases ``rate_limit``, ``rate_limiter``, ``llm_pace``).
type: object
additionalProperties: false
properties:
  max_in_flight:
    type: integer
    minimum: 0
    description: Concurrent live LLM requests per bucket. 0 disables the in-flight cap.
  requests_per_minute:
    type: number
    minimum: 0
    description: Token-bucket RPM. 0 is concurrency-only.
  keys:
    type: array
    items:
      type: string
      enum: [provider, model, user]
    description: Bucket dimensions for the live-LLM scheduler.
  user:
    type: string
  provider:
    type: string
  timeout_s:
    type: number
    minimum: 0
    description: Seconds to wait for a live slot before TimeoutError.
  state_path:
    type: [string, "null"]
    x-filesystem-path: true
    description: Shared scheduler JSON store (plus a sibling .lock file).