xiaoxia-saas/packages/shared/ai_client.py

"""豆包大模型 API 客户端（共享层）.

API 和 Worker 两边共用。基于火山引擎方舟平台的 OpenAI 兼容接口。

使用方式：
    from packages.shared.ai_client import get_doubao_client

    client = get_doubao_client()
    if client.is_available:
        result = client.chat_completion(messages=[...])
"""

from __future__ import annotations

import logging
import os
import time
import uuid
from typing import Any, Optional

import httpx

from packages.shared.config import get_shared_settings

logger = logging.getLogger(__name__)


class DoubaoClient:
    """豆包大模型 API 客户端.

    封装 OpenAI 兼容的 Chat Completion 接口，支持自动重试。
    未配置 API Key 时 is_available 为 False，调用方应降级处理。
    """

    def __init__(self) -> None:
        settings = get_shared_settings()
        self.api_key: str = settings.doubao_api_key
        self.model: str = settings.doubao_model
        self.base_url: str = settings.doubao_base_url.rstrip("/")
        self.timeout: int = settings.doubao_timeout
        self.max_retries: int = settings.doubao_max_retries
        self.vision_model: str = settings.doubao_vision_model
        self.vision_lite_model: str = settings.doubao_vision_lite_model
        self.fast_model: str = settings.doubao_fast_model

    def embed_text(self, text: str, timeout: int | None = None) -> list[float] | None:
        """调用豆包文本 Embedding API，返回浮点向量；失败返回 None。"""
        if not self.is_available or not text or not text.strip():
            return None

        url = f"{self.base_url}/embeddings"
        headers = {
            "Authorization": f"Bearer {self.api_key}",
            "Content-Type": "application/json",
        }
        payload: dict[str, Any] = {
            "model": getattr(self, "embedding_model", None) or "doubao-embedding-large-text-240915",
            "input": text.strip(),
            "encoding_format": "float",
        }

        req_timeout = timeout or self.timeout
        last_error: Exception | None = None
        for attempt in range(self.max_retries + 1):
            try:
                resp = httpx.post(url, headers=headers, json=payload, timeout=req_timeout)
                resp.raise_for_status()
                data = resp.json()
                emb_list = data.get("data") or []
                if emb_list and isinstance(emb_list, list):
                    vec = emb_list[0].get("embedding")
                    if isinstance(vec, list) and vec:
                        return [float(x) for x in vec]
                logger.warning("embedding 返回结构异常: %s", str(data)[:200])
                return None
            except Exception as e:
                last_error = e
                if attempt < self.max_retries:
                    wait = 0.5 * (2**attempt)
                    logger.warning(
                        "豆包 Embedding 调用失败，%.1fs 后重试 (%d/%d): %s", wait, attempt + 1, self.max_retries + 1, e
                    )
                    time.sleep(wait)
        logger.error("豆包 Embedding 调用最终失败: %s", last_error)
        return None

    @property
    def is_available(self) -> bool:
        """是否可用（配置了 API Key）."""
        return bool(self.api_key)

    def chat_completion(
        self,
        messages: list[dict[str, str]],
        temperature: float = 0.7,
        max_tokens: int = 1024,
        model: str | None = None,
    ) -> Optional[str]:
        """调用 Chat Completion 接口.

        Args:
            messages: 对话消息列表，[{"role": "user"/"system"/"assistant", "content": "..."}]
            temperature: 采样温度，0-2，默认0.7
            max_tokens: 最大生成token数，默认1024

        Returns:
            模型返回的文本内容，失败返回 None
        """
        if not self.is_available:
            return None

        url = f"{self.base_url}/chat/completions"
        headers = {
            "Authorization": f"Bearer {self.api_key}",
            "Content-Type": "application/json",
        }
        payload: dict[str, Any] = {
            "model": model or self.model,
            "messages": messages,
            "temperature": temperature,
            "max_tokens": max_tokens,
        }

        last_error: Optional[Exception] = None
        for attempt in range(self.max_retries + 1):
            try:
                response = httpx.post(
                    url,
                    headers=headers,
                    json=payload,
                    timeout=self.timeout,
                )
                response.raise_for_status()
                data = response.json()
                content = data["choices"][0]["message"]["content"]
                return content.strip()
            except Exception as e:
                last_error = e
                if attempt < self.max_retries:
                    wait = 0.5 * (2**attempt)
                    logger.warning(
                        "豆包API调用失败，%.1fs后重试 (第%d/%d次): %s",
                        wait,
                        attempt + 1,
                        self.max_retries + 1,
                        e,
                    )
                    time.sleep(wait)

        logger.error("豆包API调用最终失败: %s", last_error)
        return None

    def vision_completion(
        self,
        messages: list[dict],
        images: list[str] | None = None,
        max_tokens: int = 2048,
        temperature: float = 0.3,
        timeout: int | None = None,
        model: str | None = None,
    ) -> Optional[str]:
        """调用豆包视觉理解 API（OpenAI 兼容多模态格式）.

        将 images 附加到最后一条 user message 的 content 中，
        使用 vision_model（默认 doubao-1-5-vision-pro-250328）。

        Args:
            messages: 对话消息列表。最后一条 user message 会被注入图片内容。
            images: 图片列表，支持 base64 data URI 或 HTTP(S) URL。
            max_tokens: 最大生成 token 数，默认 2048。
            temperature: 采样温度，默认 0.3（视觉任务偏低更稳定）。
            timeout: 单次请求超时秒数，不传则使用默认 self.timeout。

        Returns:
            模型返回的文本内容，失败返回 None。
        """
        if not self.is_available:
            return None

        # 构造多模态 content：先追加文本，再追加图片
        vision_messages = []
        for msg in messages:
            vision_messages.append(dict(msg))

        # 将图片注入最后一条 user message
        if images and vision_messages:
            # 找到最后一条 user message
            for i in range(len(vision_messages) - 1, -1, -1):
                if vision_messages[i].get("role") == "user":
                    text_content = vision_messages[i].get("content", "")
                    multi_content: list[dict[str, Any]] = []
                    if text_content:
                        multi_content.append({"type": "text", "text": text_content})
                    for img in images:
                        if img.startswith("data:") or img.startswith("http://") or img.startswith("https://"):
                            multi_content.append({"type": "image_url", "image_url": {"url": img}})
                        else:
                            # 当作 base64 编码
                            multi_content.append(
                                {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img}"}}
                            )
                    vision_messages[i]["content"] = multi_content
                    break

        url = f"{self.base_url}/chat/completions"
        headers = {
            "Authorization": f"Bearer {self.api_key}",
            "Content-Type": "application/json",
        }
        payload: dict[str, Any] = {
            "model": model or self.vision_model,
            "messages": vision_messages,
            "temperature": temperature,
            "max_tokens": max_tokens,
        }

        req_timeout = timeout or self.timeout
        last_error: Optional[Exception] = None
        for attempt in range(self.max_retries + 1):
            try:
                response = httpx.post(
                    url,
                    headers=headers,
                    json=payload,
                    timeout=req_timeout,
                )
                response.raise_for_status()
                data = response.json()
                content = data["choices"][0]["message"]["content"]
                return content.strip()
            except Exception as e:
                last_error = e
                if attempt < self.max_retries:
                    wait = 0.5 * (2**attempt)
                    logger.warning(
                        "豆包视觉API调用失败，%.1fs后重试 (第%d/%d次): %s",
                        wait,
                        attempt + 1,
                        self.max_retries + 1,
                        e,
                    )
                    time.sleep(wait)

        logger.error("豆包视觉API调用最终失败: %s", last_error)
        return None

    # ── 视频生成（Seedance 2.5，异步任务）────────────────────────────

    def video_generation(
        self,
        prompt: str,
        *,
        image_url: str | None = None,
        duration: int = 5,
        ratio: str | None = "9:16",
        resolution: str = "720p",
        generate_audio: bool = True,
        watermark: bool = False,
        output_dir: str | None = None,
        model: str | None = None,
        reference_images: list[str] | None = None,
        reference_audios: list[str] | None = None,
        reference_videos: list[str] | None = None,
    ) -> str | None:
        """调用 Seedance 2.5 生视频（异步任务→轮询→下载），返回本地 MP4 路径；失败返回 None。

        【v1.6.1 修复】严格按官方 content 数组协议构造请求：
        - 所有参考（图/音/视）必须放进 content 数组并带 role 字段，不能放顶层 reference_audios/reference_videos（非官方字段，会被忽略或导致异常）。
        - 首帧图（first_frame 模式）Seedance 2.5 强制 ratio=adaptive；走 omni_reference（参考生视频）模式时才能指定 9:16/1:1 等具体比例。
          判定：传了参考音频/视频或 ≥1 张多参考图时，走 omni_reference（首张图 role=reference_image）；纯首帧无参考时走 first_frame（ratio 强制 adaptive）。
        - 创建任务若因 ratio 报错（HTTP 400），自动回退到 ratio=adaptive 重试一次。
        """
        if not self.is_available:
            return None
        if not prompt or not prompt.strip():
            return None

        settings = get_shared_settings()
        poll_interval = getattr(settings, "doubao_video_poll_interval", 10) or 10
        # 收紧总超时：轮询 8min + 下载 2min = 最长 ~10min，防止出现 20min 卡死
        total_timeout = getattr(settings, "doubao_video_timeout", 480) or 480
        default_video_model = getattr(settings, "doubao_video_model", None) or "doubao-seedance-2-5-260628"
        video_model = model or default_video_model

        ref_audios = [u for u in (reference_audios or [])[:10] if u and isinstance(u, str)]
        ref_videos = [u for u in (reference_videos or [])[:3] if u and isinstance(u, str)]
        ref_imgs = [u for u in (reference_images or [])[:9] if u and isinstance(u, str)]

        # 判断任务模式：有参考音/视/多图 → omni_reference（支持指定 ratio）；纯首帧 → first_frame（ratio=adaptive）
        has_extra_refs = bool(ref_audios or ref_videos or ref_imgs)
        is_first_frame_mode = bool(image_url) and not has_extra_refs
        # 最终 ratio：first_frame 模式强制 adaptive，否则按用户传值（默认 9:16）
        final_ratio = "adaptive" if is_first_frame_mode else (ratio or "9:16")

        # 构造 content 数组：text + 图 + 音 + 视
        content: list[dict[str, Any]] = [{"type": "text", "text": prompt.strip()}]
        if image_url:
            if has_extra_refs:
                # omni_reference：首张图作为 reference_image，允许指定 ratio
                content.append({
                    "type": "image_url",
                    "image_url": {"url": image_url},
                    "role": "reference_image",
                })
            else:
                # 纯首帧：不带 role，服务端识别为 first_frame（或显式 role=first_frame）
                content.append({
                    "type": "image_url",
                    "image_url": {"url": image_url},
                    "role": "first_frame",
                })
        for u in ref_imgs:
            content.append({"type": "image_url", "image_url": {"url": u}, "role": "reference_image"})
        for u in ref_audios:
            content.append({"type": "audio_url", "audio_url": {"url": u}, "role": "reference_audio"})
        for u in ref_videos:
            content.append({"type": "video_url", "video_url": {"url": u}, "role": "reference_video"})

        create_payload: dict[str, Any] = {
            "model": video_model,
            "content": content,
            "generate_audio": bool(generate_audio),
            "duration": int(duration),
            "resolution": resolution,
            "watermark": bool(watermark),
            "ratio": final_ratio,
        }

        headers = {
            "Authorization": f"Bearer {self.api_key}",
            "Content-Type": "application/json",
        }
        create_url = f"{self.base_url}/contents/generations/tasks"
        logger.info(
            "Seedance 创建任务: model=%s dur=%ds ratio=%s mode=%s gen_audio=%s img=%d aud=%d vid=%d",
            video_model,
            duration,
            final_ratio,
            "first_frame" if is_first_frame_mode else "omni_ref",
            generate_audio,
            (1 if image_url else 0) + len(ref_imgs),
            len(ref_audios),
            len(ref_videos),
        )

        def _do_create(payload: dict) -> tuple[str | None, Exception | None, int, str]:
            """返回 (task_id, last_err, status_code, body_text)。"""
            last_err: Exception | None = None
            for attempt in range(self.max_retries + 1):
                try:
                    resp = httpx.post(create_url, headers=headers, json=payload, timeout=self.timeout)
                    sc = int(getattr(resp, "status_code", 0) or 0)
                    body = (getattr(resp, "text", "") or "")[:1500]
                    if sc >= 400:
                        logger.error("Seedance 创建任务 HTTP %d: body=%s", sc, body)
                        try:
                            resp.raise_for_status()
                        except Exception as ee:
                            last_err = ee
                            if attempt < self.max_retries:
                                time.sleep(0.5 * (2 ** attempt))
                                continue
                            return None, last_err, sc, body
                    data = resp.json()
                    tid = data.get("id")
                    if tid:
                        return tid, None, sc, body
                    last_err = RuntimeError(f"create ok but no id: {str(data)[:300]}")
                except Exception as e:
                    last_err = e
                    if attempt < self.max_retries:
                        wait = 0.5 * (2 ** attempt)
                        logger.warning("Seedance 创建任务失败，%.1fs 后重试 (%d/%d): %s", wait, attempt + 1, self.max_retries + 1, e)
                        time.sleep(wait)
            return None, last_err, 0, ""

        # 第一次尝试
        task_id, last_err, sc, body = _do_create(create_payload)

        # ratio 兜底：HTTP 400 且 body 提到 ratio / adaptive → 回退 adaptive 再试一次
        if not task_id and sc == 400 and final_ratio != "adaptive" and ("ratio" in (body or "").lower() or "aspect" in (body or "").lower() or "adaptive" in (body or "").lower()):
            logger.warning("Seedance 创建因 ratio 失败，回退 ratio=adaptive 重试")
            create_payload["ratio"] = "adaptive"
            task_id, last_err, sc2, body2 = _do_create(create_payload)

        if not task_id:
            logger.error(
                "Seedance 创建任务最终失败: model=%s base_url=%s err=%s body=%s  【排查】"
                "1) 方舟控制台已开通 doubao-seedance-2-5-260628；2) API Key 有该模型权限；"
                "3) DOUBAO_BASE_URL=https://ark.cn-beijing.volces.com/api/v3；4) 参考素材 URL 公网可访问。",
                video_model, self.base_url, last_err, (body or "")[:500],
            )
            return None

        logger.info("Seedance 任务已创建: task_id=%s ratio=%s", task_id, create_payload["ratio"])

        # 2) 轮询状态
        poll_url = f"{create_url}/{task_id}"
        deadline = time.time() + total_timeout
        video_url: str | None = None
        last_status: str = "queued"
        poll_count = 0
        while time.time() < deadline:
            poll_count += 1
            try:
                resp = httpx.get(poll_url, headers=headers, timeout=self.timeout)
                try:
                    if int(getattr(resp, "status_code", 200)) >= 400:
                        resp.raise_for_status()
                except (TypeError, ValueError):
                    pass
                data = resp.json()
                status = data.get("status", "")
                last_status = status
                if status == "succeeded":
                    content_obj = data.get("content") or {}
                    video_url = content_obj.get("video_url")
                    if video_url:
                        logger.info("Seedance 任务成功: task_id=%s polls=%d", task_id, poll_count)
                        break
                    last_err = RuntimeError(f"task succeeded but no video_url: {str(data)[:300]}")
                    logger.error("Seedance succeeded 但无 video_url: %s", last_err)
                    break
                if status == "failed":
                    err = data.get("error") or {}
                    last_err = RuntimeError(f"task failed: code={err.get('code','')} msg={err.get('message','')}")
                    logger.error("Seedance 任务失败 task_id=%s: %s", task_id, last_err)
                    break
                if status in ("expired", "cancelled"):
                    last_err = RuntimeError(f"task {status}")
                    logger.error("Seedance 任务 %s: task_id=%s", status, task_id)
                    break
                # 每 5 次轮询打一次 info 日志，便于观察进度
                if poll_count % 5 == 0:
                    logger.info("Seedance 轮询中: task_id=%s status=%s polls=%d", task_id, status, poll_count)
            except httpx.HTTPStatusError as e:
                last_err = e
                logger.warning("Seedance 轮询 HTTP %d: %s", e.response.status_code, (e.response.text or "")[:300])
            except Exception as e:
                last_err = e
                logger.debug("Seedance 轮询异常: %s", e)
            time.sleep(poll_interval)

        if not video_url:
            logger.error(
                "Seedance 任务未成功: task_id=%s last_status=%s polls=%d err=%s (总等待 %.0fs)",
                task_id, last_status, poll_count, last_err, total_timeout,
            )
            return None

        # 3) 下载到本地（下载超时收紧到 120s）
        try:
            out_dir = output_dir or "/tmp"
            os.makedirs(out_dir, exist_ok=True)
            local_path = f"{out_dir}/seedance_{task_id}_{uuid.uuid4().hex[:8]}.mp4"
            download_timeout = 120.0
            logger.info("Seedance 开始下载: task_id=%s url=%s timeout=%.0fs", task_id, video_url[:120], download_timeout)
            with httpx.stream("GET", video_url, timeout=download_timeout) as r:
                r.raise_for_status()
                downloaded = 0
                with open(local_path, "wb") as f:
                    for chunk in r.iter_bytes(chunk_size=1024 * 256):
                        if chunk:
                            f.write(chunk)
                            downloaded += len(chunk)
            size = os.path.getsize(local_path)
            logger.info("Seedance 视频下载完成: %s size=%d bytes", local_path, size)
            if size == 0:
                logger.error("Seedance 下载文件大小为 0")
                try:
                    os.remove(local_path)
                except Exception:
                    pass
                return None
            return local_path
        except Exception as e:
            logger.error("Seedance 视频下载失败: %s", e, exc_info=True)
            return None


# ── 单例 ─────────────────────────────────────────────────────────────────────


_client: Optional[DoubaoClient] = None


def get_doubao_client() -> DoubaoClient:
    """获取豆包客户端单例."""
    global _client
    if _client is None:
        _client = DoubaoClient()
    return _client