Files
MoviePilot/app/schemas/event.py

358 lines
14 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
from pathlib import Path
from typing import Iterable, Optional, Dict, Any, List, Set, Callable
from pydantic import BaseModel, Field, field_validator, model_validator
from app.schemas.message import MessageChannel
from app.schemas.file import FileItem
class Event(BaseModel):
"""
事件模型
"""
event_type: str = Field(..., description="事件类型")
event_data: Optional[dict] = Field(default={}, description="事件数据")
priority: Optional[int] = Field(0, description="事件优先级")
class BaseEventData(BaseModel):
"""
事件数据的基类,所有具体事件数据类应继承自此类
"""
pass
class ConfigChangeEventData(BaseEventData):
"""
ConfigChange 事件的数据模型
"""
key: set[str] = Field(..., description="配置项的键(集合类型)")
value: Optional[Any] = Field(default=None, description="配置项的新值")
change_type: str = Field(default="update", description="配置项的变更类型,如 'add', 'update', 'delete'")
@field_validator('key', mode='before')
@classmethod
def convert_to_set(cls, v):
"""将输入的 str、list、dict.keys() 等转为 set"""
if v is None:
return set()
elif isinstance(v, str):
return {v}
elif isinstance(v, dict):
return set(str(k) for k in v.keys())
elif isinstance(v, (list, tuple)):
return set(str(item) for item in v)
elif isinstance(v, set):
return set(str(item) for item in v)
elif isinstance(v, Iterable):
return set(str(item) for item in v)
else:
return {str(v)}
class ChainEventData(BaseEventData):
"""
链式事件数据的基类,所有具体事件数据类应继承自此类
"""
pass
class AuthCredentials(ChainEventData):
"""
AuthVerification 事件的数据模型
Attributes:
username (Optional[str]): 用户名,适用于 "password" grant_type
password (Optional[str]): 用户密码,适用于 "password" grant_type
mfa_code (Optional[str]): 一次性密码,目前仅适用于 "password" 认证类型
code (Optional[str]): 授权码,适用于 "authorization_code" grant_type
grant_type (str): 认证类型,如 "password", "authorization_code", "client_credentials"
# scope (List[str]): 权限范围,如 ["read", "write"]
token (Optional[str]): 认证令牌
channel (Optional[str]): 认证渠道
service (Optional[str]): 服务名称
"""
# 输入参数
username: Optional[str] = Field(None, description="用户名,适用于 'password' 认证类型")
password: Optional[str] = Field(None, description="用户密码,适用于 'password' 认证类型")
mfa_code: Optional[str] = Field(None, description="一次性密码,目前仅适用于 'password' 认证类型")
code: Optional[str] = Field(None, description="授权码,适用于 'authorization_code' 认证类型")
grant_type: str = Field(..., description="认证类型,如 'password', 'authorization_code', 'client_credentials'")
# scope: List[str] = Field(default_factory=list, description="权限范围,如 ['read', 'write']")
# 输出参数
# grant_type 为 authorization_code 时,输出参数包括 username、token、channel、service
token: Optional[str] = Field(default=None, description="认证令牌")
channel: Optional[str] = Field(default=None, description="认证渠道")
service: Optional[str] = Field(default=None, description="服务名称")
@model_validator(mode='before')
@classmethod
def check_fields_based_on_grant_type(cls, values): # noqa
grant_type = values.get("grant_type")
if not grant_type:
values["grant_type"] = "password"
grant_type = "password"
if grant_type == "password":
if not values.get("username") or not values.get("password"):
raise ValueError("username and password are required for grant_type 'password'")
elif grant_type == "authorization_code":
if not values.get("code"):
raise ValueError("code is required for grant_type 'authorization_code'")
return values
class AuthInterceptCredentials(ChainEventData):
"""
AuthIntercept 事件的数据模型
Attributes:
# 输入参数
username (str): 用户名
channel (str): 认证渠道
service (str): 服务名称
token (str): 认证令牌
status (str): 认证状态,"triggered""completed" 两个状态
# 输出参数
source (str): 拦截源,默认值为 "未知拦截源"
cancel (bool): 是否取消认证,默认值为 False
"""
# 输入参数
username: Optional[str] = Field(..., description="用户名")
channel: str = Field(..., description="认证渠道")
service: str = Field(..., description="服务名称")
status: str = Field(..., description="认证状态, 包含 'triggered' 表示认证触发,'completed' 表示认证成功")
token: Optional[str] = Field(default=None, description="认证令牌")
# 输出参数
source: str = Field(default="未知拦截源", description="拦截源")
cancel: bool = Field(default=False, description="是否取消认证")
class CommandRegisterEventData(ChainEventData):
"""
CommandRegister 事件的数据模型
Attributes:
# 输入参数
commands (dict): 菜单命令
origin (str): 事件源,可以是 Chain 或具体的模块名称
service (str): 服务名称
# 输出参数
source (str): 拦截源,默认值为 "未知拦截源"
cancel (bool): 是否取消认证,默认值为 False
"""
# 输入参数
commands: Dict[str, dict] = Field(..., description="菜单命令")
origin: str = Field(..., description="事件源")
service: Optional[str] = Field(..., description="服务名称")
# 输出参数
cancel: bool = Field(default=False, description="是否取消注册")
source: str = Field(default="未知拦截源", description="拦截源")
class TransferRenameEventData(ChainEventData):
"""
TransferRename 事件的数据模型
Attributes:
# 输入参数
template_string (str): Jinja2 模板字符串
rename_dict (dict): 渲染上下文
render_str (str): 渲染生成的字符串
path (Optional[Path]): 当前文件的目标路径
# 输出参数
updated (bool): 是否已更新,默认值为 False
updated_str (str): 更新后的字符串
source (str): 拦截源,默认值为 "未知拦截源"
"""
# 输入参数
template_string: str = Field(..., description="模板字符串")
rename_dict: Dict[str, Any] = Field(..., description="渲染上下文")
path: Optional[Path] = Field(None, description="文件的目标路径")
render_str: str = Field(..., description="渲染生成的字符串")
# 输出参数
updated: bool = Field(default=False, description="是否已更新")
updated_str: Optional[str] = Field(default=None, description="更新后的字符串")
source: Optional[str] = Field(default="未知拦截源", description="拦截源")
class ResourceSelectionEventData(BaseModel):
"""
ResourceSelection 事件的数据模型
Attributes:
# 输入参数
contexts (List[Context]): 当前待选择的资源上下文列表
source (str): 事件源,指示事件的触发来源
# 输出参数
updated (bool): 是否已更新,默认值为 False
updated_contexts (Optional[List[Context]]): 已更新的资源上下文列表,默认值为 None
source (str): 更新源,默认值为 "未知更新源"
"""
# 输入参数
contexts: Any = Field(None, description="待选择的资源上下文列表")
downloader: Optional[str] = Field(None, description="下载器")
origin: Optional[str] = Field(None, description="来源")
# 输出参数
updated: bool = Field(default=False, description="是否已更新")
updated_contexts: Optional[List[Any]] = Field(default=None, description="已更新的资源上下文列表")
source: Optional[str] = Field(default="未知拦截源", description="拦截源")
class ResourceDownloadEventData(ChainEventData):
"""
ResourceDownload 事件的数据模型
Attributes:
# 输入参数
context (Context): 当前资源上下文
episodes (Set[int]): 需要下载的集数
channel (MessageChannel): 通知渠道
origin (str): 来源消息通知、Subscribe、Manual等
downloader (str): 下载器
options (dict): 其他参数
# 输出参数
cancel (bool): 是否取消下载,默认值为 False
source (str): 拦截源,默认值为 "未知拦截源"
reason (str): 拦截原因,描述拦截的具体原因
"""
# 输入参数
context: Any = Field(None, description="当前资源上下文")
episodes: Optional[Set[int]] = Field(None, description="需要下载的集数")
channel: Optional[MessageChannel] = Field(None, description="通知渠道")
origin: Optional[str] = Field(None, description="来源")
downloader: Optional[str] = Field(None, description="下载器")
options: Optional[dict] = Field(default={}, description="其他参数")
# 输出参数
cancel: bool = Field(default=False, description="是否取消下载")
source: str = Field(default="未知拦截源", description="拦截源")
reason: str = Field(default="", description="拦截原因")
class TransferInterceptEventData(ChainEventData):
"""
TransferIntercept 事件的数据模型
Attributes:
# 输入参数
fileitem (FileItem): 源文件
target_storage (str): 目标存储
target_path (Path): 目标路径
transfer_type (str): 整理方式copy、move、link、softlink等
options (dict): 其他参数
# 输出参数
cancel (bool): 是否取消下载,默认值为 False
source (str): 拦截源,默认值为 "未知拦截源"
reason (str): 拦截原因,描述拦截的具体原因
"""
# 输入参数
fileitem: FileItem = Field(..., description="源文件")
mediainfo: Any = Field(..., description="媒体信息")
target_storage: str = Field(..., description="目标存储")
target_path: Path = Field(..., description="目标路径")
transfer_type: str = Field(..., description="整理方式")
options: Optional[dict] = Field(default=None, description="其他参数")
# 输出参数
cancel: bool = Field(default=False, description="是否取消整理")
source: str = Field(default="未知拦截源", description="拦截源")
reason: str = Field(default="", description="拦截原因")
class DiscoverMediaSource(BaseModel):
"""
探索媒体数据源的基类
"""
name: str = Field(..., description="数据源名称")
mediaid_prefix: str = Field(..., description="媒体ID的前缀不含:")
api_path: str = Field(..., description="媒体数据源API地址")
filter_params: Optional[Dict[str, Any]] = Field(default=None, description="过滤参数")
filter_ui: Optional[List[dict]] = Field(default=[], description="过滤参数UI配置")
depends: Optional[Dict[str, list]] = Field(default=None, description="UI依赖关系字典")
class DiscoverSourceEventData(ChainEventData):
"""
DiscoverSource 事件的数据模型
Attributes:
# 输出参数
extra_sources (List[DiscoverMediaSource]): 额外媒体数据源
"""
# 输出参数
extra_sources: List[DiscoverMediaSource] = Field(default_factory=list, description="额外媒体数据源")
class RecommendMediaSource(BaseModel):
"""
推荐媒体数据源的基类
"""
name: str = Field(..., description="数据源名称")
api_path: str = Field(..., description="媒体数据源API地址")
type: str = Field(..., description="类型")
class RecommendSourceEventData(ChainEventData):
"""
RecommendSource 事件的数据模型
Attributes:
# 输出参数
extra_sources (List[RecommendMediaSource]): 额外媒体数据源
"""
# 输出参数
extra_sources: List[RecommendMediaSource] = Field(default_factory=list, description="额外媒体数据源")
class MediaRecognizeConvertEventData(ChainEventData):
"""
MediaRecognizeConvert 事件的数据模型
Attributes:
# 输入参数
mediaid (str): 媒体ID格式为`前缀:ID值`,如 tmdb:12345、douban:1234567
convert_type (str): 转换类型 仅支持themoviedb/douban需要转换为对应的媒体数据并返回
# 输出参数
media_dict (dict): TheMovieDb/豆瓣的媒体数据
"""
# 输入参数
mediaid: str = Field(..., description="媒体ID")
convert_type: str = Field(..., description="转换类型themoviedb/douban")
# 输出参数
media_dict: dict = Field(default_factory=dict, description="转换后的媒体信息TheMovieDb/豆瓣)")
class StorageOperSelectionEventData(ChainEventData):
"""
StorageOperSelect 事件的数据模型
Attributes:
# 输入参数
storage (str): 存储类型
# 输出参数
storage_oper (Callable): 存储操作对象
"""
# 输入参数
storage: Optional[str] = Field(default=None, description="存储类型")
# 输出参数
storage_oper: Optional[Callable] = Field(default=None, description="存储操作对象")