dsh-plugin-image-workbench 0.0.0-stage → 1.1.5

package/vendor/chatgpt2api/source/services/image_task_service.py

@@ -0,0 +1,550 @@
+from __future__ import annotations
+
+import json
+import threading
+import time
+from collections.abc import Callable
+from datetime import datetime
+from pathlib import Path
+from typing import Any
+
+from services.config import DATA_DIR, config
+from services.content_filter import request_text
+from services.log_service import LOG_TYPE_CALL, log_service
+from services.protocol import openai_v1_image_edit, openai_v1_image_generations
+
+TASK_STATUS_QUEUED = "queued"
+TASK_STATUS_RUNNING = "running"
+TASK_STATUS_SUCCESS = "success"
+TASK_STATUS_ERROR = "error"
+TERMINAL_STATUSES = {TASK_STATUS_SUCCESS, TASK_STATUS_ERROR}
+UNFINISHED_STATUSES = {TASK_STATUS_QUEUED, TASK_STATUS_RUNNING}
+
+
+def _now_iso() -> str:
+    return datetime.now().strftime("%Y-%m-%d %H:%M:%S")
+
+
+def _timestamp(value: object) -> float:
+    if not isinstance(value, str) or not value.strip():
+        return 0.0
+    for fmt in ("%Y-%m-%d %H:%M:%S", "%Y-%m-%dT%H:%M:%S.%f", "%Y-%m-%dT%H:%M:%S"):
+        try:
+            return datetime.strptime(value[:26], fmt).timestamp()
+        except ValueError:
+            continue
+    try:
+        return datetime.fromisoformat(value.replace("Z", "+00:00")).timestamp()
+    except Exception:
+        return 0.0
+
+
+def _clean(value: object, default: str = "") -> str:
+    return str(value or default).strip()
+
+
+def _owner_id(identity: dict[str, object]) -> str:
+    return _clean(identity.get("id")) or "anonymous"
+
+
+def _task_key(owner_id: str, task_id: str) -> str:
+    return f"{owner_id}:{task_id}"
+
+
+def _collect_image_urls(data: list[Any]) -> list[str]:
+    urls: list[str] = []
+    for item in data:
+        if isinstance(item, dict):
+            url = item.get("url")
+            if isinstance(url, str) and url:
+                urls.append(url)
+    return urls
+
+
+def _public_task(task: dict[str, Any]) -> dict[str, Any]:
+    item = {
+        "id": task.get("id"),
+        "status": task.get("status"),
+        "mode": task.get("mode"),
+        "model": task.get("model"),
+        "size": task.get("size"),
+        "quality": task.get("quality"),
+        "created_at": task.get("created_at"),
+        "updated_at": task.get("updated_at"),
+    }
+    if task.get("conversation_id"):
+        item["conversation_id"] = task.get("conversation_id")
+    if task.get("data") is not None:
+        item["data"] = task.get("data")
+    if task.get("usage") is not None:
+        item["usage"] = task.get("usage")
+    if task.get("error"):
+        item["error"] = task.get("error")
+    if task.get("progress"):
+        item["progress"] = task.get("progress")
+    if task.get("duration_ms") is not None:
+        item["duration_ms"] = task.get("duration_ms")
+    if task.get("status") in (TASK_STATUS_RUNNING, TASK_STATUS_QUEUED):
+        if task.get("status") == TASK_STATUS_RUNNING:
+            # RUNNING 状态仅在 started_ts 被设置后（image_stream_resolve_start）才计时
+            base_ts = task.get("started_ts")
+        else:
+            # QUEUED 状态从 created_ts 开始计时（排队等待中）
+            base_ts = task.get("created_ts") or task.get("updated_ts")
+        if base_ts:
+            item["elapsed_secs"] = round(time.time() - base_ts, 1)
+    return item
+
+
+class ImageTaskService:
+    def __init__(
+        self,
+        path: Path,
+        *,
+        generation_handler: Callable[[dict[str, Any]], dict[str, Any]] = openai_v1_image_generations.handle,
+        edit_handler: Callable[[dict[str, Any]], dict[str, Any]] = openai_v1_image_edit.handle,
+        retention_days_getter: Callable[[], int] | None = None,
+    ):
+        self.path = path
+        self.generation_handler = generation_handler
+        self.edit_handler = edit_handler
+        self.retention_days_getter = retention_days_getter or (lambda: config.image_retention_days)
+        self._lock = threading.RLock()
+        self._tasks: dict[str, dict[str, Any]] = {}
+        self.path.parent.mkdir(parents=True, exist_ok=True)
+        with self._lock:
+            self._tasks = self._load_locked()
+            changed = self._recover_unfinished_locked()
+            changed = self._cleanup_locked() or changed
+            if changed:
+                self._save_locked()
+
+    def submit_generation(
+        self,
+        identity: dict[str, object],
+        *,
+        client_task_id: str,
+        prompt: str,
+        model: str,
+        size: str | None,
+        quality: str = "auto",
+        base_url: str = "",
+    ) -> dict[str, Any]:
+        payload = {
+            "prompt": prompt,
+            "model": model,
+            "n": 1,
+            "size": size,
+            "quality": quality,
+            "response_format": "url",
+            "base_url": base_url,
+        }
+        return self._submit(identity, client_task_id=client_task_id, mode="generate", payload=payload)
+
+    def submit_edit(
+        self,
+        identity: dict[str, object],
+        *,
+        client_task_id: str,
+        prompt: str,
+        model: str,
+        size: str | None,
+        quality: str = "auto",
+        base_url: str = "",
+        images: list[tuple[bytes, str, str]] | None = None,
+        masks: list[tuple[bytes, str, str]] | None = None,
+    ) -> dict[str, Any]:
+        payload = {
+            "prompt": prompt,
+            "images": images or [],
+            "mask": masks or [],
+            "model": model,
+            "n": 1,
+            "size": size,
+            "quality": quality,
+            "response_format": "url",
+            "base_url": base_url,
+        }
+        return self._submit(identity, client_task_id=client_task_id, mode="edit", payload=payload)
+
+    def list_tasks(self, identity: dict[str, object], task_ids: list[str]) -> dict[str, Any]:
+        owner = _owner_id(identity)
+        requested_ids = [_clean(task_id) for task_id in task_ids if _clean(task_id)]
+        with self._lock:
+            if self._cleanup_locked():
+                self._save_locked()
+            items = []
+            missing_ids = []
+            for task_id in requested_ids:
+                task = self._tasks.get(_task_key(owner, task_id))
+                if task is None:
+                    missing_ids.append(task_id)
+                else:
+                    items.append(_public_task(task))
+            if not requested_ids:
+                items = [
+                    _public_task(task)
+                    for task in self._tasks.values()
+                    if task.get("owner_id") == owner
+                ]
+                items.sort(key=lambda item: str(item.get("updated_at") or ""), reverse=True)
+                missing_ids = []
+            return {"items": items, "missing_ids": missing_ids}
+
+    def _submit(
+        self,
+        identity: dict[str, object],
+        *,
+        client_task_id: str,
+        mode: str,
+        payload: dict[str, Any],
+    ) -> dict[str, Any]:
+        task_id = _clean(client_task_id)
+        if not task_id:
+            raise ValueError("client_task_id is required")
+        owner = _owner_id(identity)
+        key = _task_key(owner, task_id)
+        now = _now_iso()
+        should_start = False
+        with self._lock:
+            cleaned = self._cleanup_locked()
+            task = self._tasks.get(key)
+            if task is not None:
+                if cleaned:
+                    self._save_locked()
+                return _public_task(task)
+            task = {
+                "id": task_id,
+                "owner_id": owner,
+                "status": TASK_STATUS_QUEUED,
+                "mode": mode,
+                "model": _clean(payload.get("model"), "gpt-image-2"),
+                "size": _clean(payload.get("size")),
+                "quality": _clean(payload.get("quality"), "auto"),
+                "created_at": now,
+                "updated_at": now,
+                "created_ts": time.time(),
+            }
+            self._tasks[key] = task
+            self._save_locked()
+            should_start = True
+
+        if should_start:
+            thread = threading.Thread(
+                target=self._run_task,
+                args=(key, mode, payload, dict(identity), _clean(payload.get("model"), "gpt-image-2")),
+                name=f"image-task-{task_id[:16]}",
+                daemon=True,
+            )
+            thread.start()
+        return _public_task(task)
+
+    def _run_task(
+        self,
+        key: str,
+        mode: str,
+        payload: dict[str, Any],
+        identity: dict[str, object],
+        model: str,
+    ) -> None:
+        started = time.time()
+        self._update_task(key, status=TASK_STATUS_RUNNING, error="")
+        # 创建进度回调，每个步骤完成后更新任务状态
+        def progress_callback(step: str) -> None:
+            if step == "image_stream_resolve_start":
+                self._update_task(key, started_ts=time.time())
+            self._update_task(key, progress=step)
+        # 将进度回调添加到 payload 中（handler 会提取并传递给 ConversationRequest）
+        payload_with_progress = {**payload, "progress_callback": progress_callback}
+        try:
+            handler = self.edit_handler if mode == "edit" else self.generation_handler
+            result = handler(payload_with_progress)
+            if not isinstance(result, dict):
+                raise RuntimeError("image task returned streaming result unexpectedly")
+            data = result.get("data")
+            account_email = _clean(result.get("_account_email") or result.get("account_email"))
+            if not isinstance(data, list) or not data:
+                upstream = _clean(result.get("message"))
+                if upstream:
+                    message = upstream
+                else:
+                    message = "号池中没有可用账号或所有账号均被限流，请检查号池状态（账号额度、是否被封禁、是否到达生图上限）"
+                error = RuntimeError(message)
+                if account_email:
+                    setattr(error, "account_email", account_email)
+                raise error
+            usage = result.get("usage")
+            duration_ms = int((time.time() - started) * 1000)
+            self._update_task(key, status=TASK_STATUS_SUCCESS, data=data, usage=usage, error="", duration_ms=duration_ms)
+            self._log_call(
+                identity,
+                mode,
+                model,
+                started,
+                "调用完成",
+                request_preview=request_text(payload.get("prompt")),
+                urls=_collect_image_urls(data),
+                account_email=account_email,
+            )
+        except Exception as exc:
+            error_message = str(exc) or "image task failed"
+            account_email = _clean(getattr(exc, "account_email", ""))
+            conversation_id = _clean(getattr(exc, "conversation_id", ""))
+            duration_ms = int((time.time() - started) * 1000)
+            self._update_task(key, status=TASK_STATUS_ERROR, error=error_message, data=[],
+                              duration_ms=duration_ms,
+                              **({"conversation_id": conversation_id} if conversation_id else {}))
+            self._log_call(
+                identity,
+                mode,
+                model,
+                started,
+                "调用失败",
+                request_preview=request_text(payload.get("prompt")),
+                status="failed",
+                error=error_message,
+                account_email=account_email,
+            )
+
+    def _log_call(
+        self,
+        identity: dict[str, object],
+        mode: str,
+        model: str,
+        started: float,
+        suffix: str,
+        *,
+        request_preview: str = "",
+        status: str = "success",
+        error: str = "",
+        urls: list[str] | None = None,
+        account_email: str = "",
+    ) -> None:
+        endpoint = "/v1/images/edits" if mode == "edit" else "/v1/images/generations"
+        summary_prefix = "图生图" if mode == "edit" else "文生图"
+        detail = {
+            "key_id": identity.get("id"),
+            "key_name": identity.get("name"),
+            "role": identity.get("role"),
+            "endpoint": endpoint,
+            "model": model,
+            "started_at": datetime.fromtimestamp(started).strftime("%Y-%m-%d %H:%M:%S"),
+            "ended_at": _now_iso(),
+            "duration_ms": int((time.time() - started) * 1000),
+            "status": status,
+        }
+        if request_preview:
+            detail["request_text"] = request_preview
+        if error:
+            detail["error"] = error
+        if account_email:
+            detail["account_email"] = account_email
+        if urls:
+            detail["urls"] = list(dict.fromkeys(urls))
+        try:
+            log_service.add(LOG_TYPE_CALL, f"{summary_prefix}{suffix}", detail)
+        except Exception:
+            pass
+
+    def _update_task(self, key: str, **updates: Any) -> None:
+        with self._lock:
+            task = self._tasks.get(key)
+            if task is None:
+                return
+            task.update(updates)
+            task["updated_at"] = _now_iso()
+            task["updated_ts"] = time.time()
+            self._save_locked()
+
+    def _load_locked(self) -> dict[str, dict[str, Any]]:
+        if not self.path.exists():
+            return {}
+        try:
+            raw = json.loads(self.path.read_text(encoding="utf-8"))
+        except Exception:
+            return {}
+        raw_items = raw.get("tasks") if isinstance(raw, dict) else raw
+        if not isinstance(raw_items, list):
+            return {}
+        tasks: dict[str, dict[str, Any]] = {}
+        for item in raw_items:
+            if not isinstance(item, dict):
+                continue
+            task_id = _clean(item.get("id"))
+            owner = _clean(item.get("owner_id"))
+            if not task_id or not owner:
+                continue
+            status = _clean(item.get("status"))
+            if status not in {TASK_STATUS_QUEUED, TASK_STATUS_RUNNING, TASK_STATUS_SUCCESS, TASK_STATUS_ERROR}:
+                status = TASK_STATUS_ERROR
+            task = {
+                "id": task_id,
+                "owner_id": owner,
+                "status": status,
+                "mode": "edit" if item.get("mode") == "edit" else "generate",
+                "model": _clean(item.get("model"), "gpt-image-2"),
+                "size": _clean(item.get("size")),
+                "quality": _clean(item.get("quality"), "auto"),
+                "created_at": _clean(item.get("created_at"), _now_iso()),
+                "updated_at": _clean(item.get("updated_at"), _clean(item.get("created_at"), _now_iso())),
+                "created_ts": item.get("created_ts"),
+                "updated_ts": item.get("updated_ts"),
+                "started_ts": item.get("started_ts"),
+                "duration_ms": item.get("duration_ms"),
+            }
+            data = item.get("data")
+            if isinstance(data, list):
+                task["data"] = data
+            usage = item.get("usage")
+            if isinstance(usage, dict):
+                task["usage"] = usage
+            error = _clean(item.get("error"))
+            if error:
+                task["error"] = error
+            tasks[_task_key(owner, task_id)] = task
+        return tasks
+
+    def _save_locked(self) -> None:
+        items = sorted(self._tasks.values(), key=lambda item: str(item.get("updated_at") or ""), reverse=True)
+        tmp_path = self.path.with_suffix(self.path.suffix + ".tmp")
+        tmp_path.write_text(json.dumps({"tasks": items}, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
+        tmp_path.replace(self.path)
+
+    def _recover_unfinished_locked(self) -> bool:
+        changed = False
+        for task in self._tasks.values():
+            if task.get("status") in UNFINISHED_STATUSES:
+                task["status"] = TASK_STATUS_ERROR
+                task["error"] = "服务已重启，未完成的图片任务已中断"
+                task["updated_at"] = _now_iso()
+                changed = True
+        return changed
+
+    def _cleanup_locked(self) -> bool:
+        try:
+            retention_days = max(1, int(self.retention_days_getter()))
+        except Exception:
+            retention_days = 30
+        cutoff = time.time() - retention_days * 86400
+        removed_keys = [
+            key
+            for key, task in self._tasks.items()
+            if task.get("status") in TERMINAL_STATUSES and _timestamp(task.get("updated_at")) < cutoff
+        ]
+        for key in removed_keys:
+            self._tasks.pop(key, None)
+        return bool(removed_keys)
+
+    def resume_poll(
+        self,
+        identity: dict[str, object],
+        task_id: str,
+        extra_timeout_secs: float = 30.0,
+    ) -> dict[str, Any]:
+        """恢复对已超时任务的轮询，额外等待 extra_timeout_secs 秒。"""
+        owner = _owner_id(identity)
+        key = _task_key(owner, _clean(task_id))
+        with self._lock:
+            task = self._tasks.get(key)
+            if task is None:
+                raise ValueError("task not found")
+            if task.get("status") != TASK_STATUS_ERROR:
+                raise ValueError("task is not in error state")
+            error_msg = _clean(task.get("error"))
+            if "超时" not in error_msg:
+                raise ValueError("task error is not a timeout error")
+            conversation_id = _clean(task.get("conversation_id"))
+            if not conversation_id:
+                raise ValueError("task has no conversation_id")
+            mode = task.get("mode", "generate")
+            model = task.get("model", "gpt-image-2")
+            # 将任务状态重置为 running
+            self._update_task(key, status=TASK_STATUS_RUNNING, error="")
+
+        # 启动新线程继续轮询
+        thread = threading.Thread(
+            target=self._run_resume_poll,
+            args=(key, conversation_id, extra_timeout_secs, dict(identity), mode, model),
+            name=f"image-resume-{_clean(task_id)[:16]}",
+            daemon=True,
+        )
+        thread.start()
+        return _public_task(task)
+
+    def _run_resume_poll(
+        self,
+        key: str,
+        conversation_id: str,
+        extra_timeout_secs: float,
+        identity: dict[str, object],
+        mode: str,
+        model: str,
+    ) -> None:
+        """后台线程：继续轮询已有 conversation_id 的图片结果。"""
+        started = time.time()
+        backend = None
+        try:
+            from services.openai_backend_api import OpenAIBackendAPI
+            from services.protocol.conversation import format_image_result
+
+            backend = OpenAIBackendAPI(proxy_url=config.proxy_url or None)
+            file_ids, sediment_ids = backend._poll_image_results(
+                conversation_id,
+                extra_timeout_secs,
+            )
+            if not file_ids and not sediment_ids:
+                raise RuntimeError(
+                    f"继续等待 {extra_timeout_secs} 秒后仍未找到图片结果。"
+                )
+
+            image_urls = backend.resolve_conversation_image_urls(
+                conversation_id, file_ids, sediment_ids, poll=False,
+            )
+            if not image_urls:
+                raise RuntimeError("图片 URL 解析失败")
+
+            image_items = [
+                {"b64_json": __import__("base64").b64encode(image_data).decode("ascii")}
+                for image_data in backend.download_image_bytes(image_urls)
+            ]
+            # 获取 task 的原始 prompt（从 _public_task 的 mode 判断）
+            with self._lock:
+                task = self._tasks.get(key)
+                quality = _clean(task.get("quality"), "auto") if task else "auto"
+                size = _clean(task.get("size")) if task else None
+            data = format_image_result(
+                image_items,
+                "",  # prompt 已不重要，结果已经拿到了
+                "b64_json",
+                "",
+                int(time.time()),
+            )["data"]
+            self._update_task(key, status=TASK_STATUS_SUCCESS, data=data, error="", duration_ms=int((time.time() - started) * 1000))
+            self._log_call(
+                identity,
+                mode,
+                model,
+                started,
+                "调用完成（续轮询）",
+                status="success",
+                urls=_collect_image_urls(data),
+            )
+        except Exception as exc:
+            error_message = str(exc) or "resume poll failed"
+            duration_ms = int((time.time() - started) * 1000)
+            self._update_task(key, status=TASK_STATUS_ERROR, error=error_message, data=[], duration_ms=duration_ms)
+            self._log_call(
+                identity,
+                mode,
+                model,
+                started,
+                "调用失败（续轮询）",
+                status="failed",
+                error=error_message,
+            )
+        finally:
+            if backend is not None:
+                backend.close()
+
+
+image_task_service = ImageTaskService(DATA_DIR / "image_tasks.json")