"""LLM answer generation with the客服 Agent fallback chain."""
from __future__ import annotations

import logging
from inspect import isawaitable


logger = logging.getLogger("rag.generation")


async def _config(config_getter, key: str, default=None):
    value = config_getter(key, default)
    if isawaitable(value):
        value = await value
    return value


async def generate_answer(
    messages: list[dict],
    *,
    llm_client,
    config_getter,
    primary_model: str | None = None,
) -> str:
    fallback_model = await _config(
        config_getter, "agent.customer.llm.fallback_model", ""
    )
    template = await _config(
        config_getter, "agent.customer.template.system_busy", None
    )
    models = [primary_model] if primary_model else [None]
    if fallback_model and fallback_model not in models:
        models.append(fallback_model)
    for model in models:
        try:
            kwargs = {} if model is None else {"model": model}
            return await llm_client.chat(messages, **kwargs)
        except Exception:
            logger.exception("LLM generation failed for model=%s", model or "default")
    if not template:
        raise RuntimeError("system busy template is not configured")
    return template