deepdata: core product code (backend, frontend, runtime, deepflow module)
This commit is contained in:
@@ -0,0 +1,177 @@
|
||||
import uuid
|
||||
|
||||
from sqlalchemy import Column, String, Text, Integer, JSON, TIMESTAMP, func
|
||||
|
||||
from app.db.models.base_entity import Base, BaseEntity
|
||||
from app.module.generation.schema.generation import CreateSynthesisTaskRequest
|
||||
|
||||
|
||||
async def save_synthesis_task(db_session, synthesis_task: CreateSynthesisTaskRequest):
|
||||
"""保存数据合成任务。
|
||||
|
||||
注意:当前 MySQL 表 `t_data_synth_instances` 结构中只包含 synth_type / synth_config 等字段,
|
||||
没有 model_id、text_split_config、source_file_id、result_data_location 等列,因此这里只保存
|
||||
与表结构一致的字段,其他信息由上层逻辑或其它表负责管理。
|
||||
"""
|
||||
gid = str(uuid.uuid4())
|
||||
|
||||
# 兼容旧请求结构:从请求对象中提取必要字段,
|
||||
# - 合成类型:synthesis_type -> synth_type
|
||||
# - 合成配置:text_split_config + synthesis_config 合并后写入 synth_config
|
||||
|
||||
synth_task_instance = DataSynthInstance(
|
||||
id=gid,
|
||||
name=synthesis_task.name,
|
||||
description=synthesis_task.description,
|
||||
status="pending",
|
||||
synth_type=synthesis_task.synthesis_type.value,
|
||||
progress=0,
|
||||
synth_config=synthesis_task.synth_config.model_dump(),
|
||||
total_files=len(synthesis_task.source_file_id or []),
|
||||
processed_files=0,
|
||||
total_chunks=0,
|
||||
processed_chunks=0,
|
||||
total_synth_data=0,
|
||||
created_at=func.now(),
|
||||
updated_at=func.now(),
|
||||
created_by="system",
|
||||
updated_by="system",
|
||||
)
|
||||
db_session.add(synth_task_instance)
|
||||
await db_session.commit()
|
||||
await db_session.refresh(synth_task_instance)
|
||||
return synth_task_instance
|
||||
|
||||
|
||||
class DataSynthInstance(BaseEntity):
|
||||
"""数据合成任务表,对应表 t_data_synth_instances
|
||||
|
||||
create table if not exists t_data_synth_instances
|
||||
(
|
||||
id VARCHAR(36) CHARACTER SET utf8mb4 COLLATE utf8mb4_unicode_ci PRIMARY KEY COMMENT 'UUID',
|
||||
name VARCHAR(255) NOT NULL COMMENT '任务名称',
|
||||
description TEXT COMMENT '任务描述',
|
||||
status VARCHAR(20) COMMENT '任务状态',
|
||||
synth_type VARCHAR(20) NOT NULL COMMENT '合成类型',
|
||||
progress INT DEFAULT 0 COMMENT '任务进度(百分比)',
|
||||
synth_config JSON NOT NULL COMMENT '合成配置',
|
||||
total_files INT DEFAULT 0 COMMENT '总文件数',
|
||||
processed_files INT DEFAULT 0 COMMENT '已处理文件数',
|
||||
total_chunks INT DEFAULT 0 COMMENT '总文本块数',
|
||||
processed_chunks INT DEFAULT 0 COMMENT '已处理文本块数',
|
||||
total_synth_data INT DEFAULT 0 COMMENT '总合成数据量',
|
||||
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP COMMENT '创建时间',
|
||||
updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP COMMENT '更新时间',
|
||||
created_by VARCHAR(255) COMMENT '创建者',
|
||||
updated_by VARCHAR(255) COMMENT '更新者'
|
||||
) COMMENT='数据合成任务表(UUID 主键)';
|
||||
"""
|
||||
|
||||
__tablename__ = "t_data_synth_instances"
|
||||
|
||||
id = Column(String(36), primary_key=True, index=True, comment="UUID")
|
||||
name = Column(String(255), nullable=False, comment="任务名称")
|
||||
description = Column(Text, nullable=True, comment="任务描述")
|
||||
status = Column(String(20), nullable=True, comment="任务状态")
|
||||
# 与数据库字段保持一致:synth_type / synth_config
|
||||
synth_type = Column(String(20), nullable=False, comment="合成类型")
|
||||
progress = Column(Integer, nullable=False, default=0, comment="任务进度(百分比)")
|
||||
synth_config = Column(JSON, nullable=False, comment="合成配置")
|
||||
total_files = Column(Integer, nullable=False, default=0, comment="总文件数")
|
||||
processed_files = Column(Integer, nullable=False, default=0, comment="已处理文件数")
|
||||
total_chunks = Column(Integer, nullable=False, default=0, comment="总文本块数")
|
||||
processed_chunks = Column(Integer, nullable=False, default=0, comment="已处理文本块数")
|
||||
total_synth_data = Column(Integer, nullable=False, default=0, comment="总合成数据量")
|
||||
|
||||
|
||||
class DataSynthesisFileInstance(BaseEntity):
|
||||
"""数据合成文件任务表,对应表 t_data_synthesis_file_instances
|
||||
|
||||
create table if not exists t_data_synthesis_file_instances (
|
||||
id VARCHAR(36) PRIMARY KEY COMMENT 'UUID',
|
||||
synthesis_instance_id VARCHAR(36) COMMENT '数据合成任务ID',
|
||||
file_name VARCHAR(255) NOT NULL COMMENT '文件名',
|
||||
source_file_id VARCHAR(255) NOT NULL COMMENT '原始文件ID',
|
||||
target_file_location VARCHAR(1000) NOT NULL COMMENT '目标文件存储位置',
|
||||
status VARCHAR(20) COMMENT '任务状态',
|
||||
total_chunks INT DEFAULT 0 COMMENT '总文本块数',
|
||||
processed_chunks INT DEFAULT 0 COMMENT '已处理文本块数',
|
||||
created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP COMMENT '创建时间',
|
||||
updated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP ON UPDATE CURRENT_TIMESTAMP COMMENT '更新时间',
|
||||
created_by VARCHAR(255) COMMENT '创建者',
|
||||
updated_by VARCHAR(255) COMMENT '更新者'
|
||||
) COMMENT='数据合成文件任务表(UUID 主键)';
|
||||
"""
|
||||
|
||||
__tablename__ = "t_data_synthesis_file_instances"
|
||||
|
||||
id = Column(String(36), primary_key=True, index=True, comment="UUID")
|
||||
synthesis_instance_id = Column(
|
||||
String(36),
|
||||
nullable=False,
|
||||
comment="数据合成任务ID",
|
||||
index=True,
|
||||
)
|
||||
file_name = Column(String(255), nullable=False, comment="文件名")
|
||||
source_file_id = Column(String(255), nullable=False, comment="原始文件ID")
|
||||
target_file_location = Column(String(1000), nullable=True, comment="目标文件存储位置")
|
||||
status = Column(String(20), nullable=True, comment="任务状态")
|
||||
total_chunks = Column(Integer, nullable=False, default=0, comment="总文本块数")
|
||||
processed_chunks = Column(Integer, nullable=False, default=0, comment="已处理文本块数")
|
||||
|
||||
|
||||
class DataSynthesisChunkInstance(Base):
|
||||
"""数据合成分块任务表,对应表 t_data_synthesis_chunk_instances
|
||||
|
||||
create table if not exists t_data_synthesis_chunk_instances (
|
||||
id VARCHAR(36) PRIMARY KEY COMMENT 'UUID',
|
||||
synthesis_file_instance_id VARCHAR(36) COMMENT '数据合成文件任务ID',
|
||||
chunk_index INT COMMENT '分块索引',
|
||||
chunk_content TEXT COMMENT '分块内容',
|
||||
metadata JSON COMMENT '分块元数据'
|
||||
) COMMENT='数据合成分块任务表(UUID 主键)';
|
||||
"""
|
||||
|
||||
__tablename__ = "t_data_synthesis_chunk_instances"
|
||||
|
||||
id = Column(String(36), primary_key=True, index=True, comment="UUID")
|
||||
synthesis_file_instance_id = Column(
|
||||
String(36),
|
||||
nullable=False,
|
||||
comment="数据合成文件任务ID",
|
||||
index=True,
|
||||
)
|
||||
chunk_index = Column(Integer, nullable=True, comment="分块索引")
|
||||
chunk_content = Column(Text, nullable=True, comment="分块内容")
|
||||
# SQLAlchemy Declarative 保留了属性名 'metadata',这里使用 chunk_metadata 作为属性名,
|
||||
# 底层列名仍为 'metadata' 以保持与表结构兼容。
|
||||
chunk_metadata = Column("metadata", JSON, nullable=True, comment="分块元数据")
|
||||
|
||||
|
||||
class SynthesisData(Base):
|
||||
"""数据合成结果表,对应表 t_synthesis_data
|
||||
|
||||
create table if not exists t_synthesis_data (
|
||||
id VARCHAR(36) PRIMARY KEY COMMENT 'UUID',
|
||||
data json COMMENT '合成的数据',
|
||||
synthesis_file_instance_id VARCHAR(36) COMMENT '数据合成文件任务ID',
|
||||
chunk_instance_id VARCHAR(36) COMMENT '分块任务ID'
|
||||
) COMMENT='数据合成任务队列表(UUID 主键)';
|
||||
"""
|
||||
|
||||
__tablename__ = "t_data_synthesis_data"
|
||||
|
||||
id = Column(String(36), primary_key=True, index=True, comment="UUID")
|
||||
data = Column(JSON, nullable=True, comment="合成的数据")
|
||||
synthesis_file_instance_id = Column(
|
||||
String(36),
|
||||
nullable=False,
|
||||
comment="数据合成文件任务ID",
|
||||
index=True,
|
||||
)
|
||||
chunk_instance_id = Column(
|
||||
String(36),
|
||||
nullable=False,
|
||||
comment="分块任务ID",
|
||||
index=True,
|
||||
)
|
||||
Reference in New Issue
Block a user