mirror of
https://github.com/ThearchyHelios/AgentLab.git
synced 2026-11-15 06:02:01 +00:00
ThearchyHelios
8d9a712eae
- 新增标准业务实体（CanonicalEntity）模型、确定性跨源编码翻译、后台候选实体扫描与文档驱动实体智能助手 - 新增指标本体图谱（factor / reconciles_with / upstream_driver）、DAG 防环校验以及 ontology_neighbors / source_health 内置工具 - 新增跨源对账（analysis_reconcile）与滞后共变（analysis_cochange）分析算子，打通 [[h:H1]] 数据源健康证据链、报告导出与本体消融评测 - 数据目录新增「标准业务实体」视图与候选审阅工作台，口径页集成指标本体拓扑面板与端到端浏览器检查
335 lines
11 KiB
Python
335 lines
11 KiB
Python
"""标准业务实体维护的业务文档与主数据对照表附件服务（E1）。
|
||
|
||
支持两类文档来源：
|
||
1. 临时上传附件（POST /api/ontology/entities/attachments）：复用 memory/parsing.extract(..., tables=True)
|
||
提取 .md / .txt / .docx / .pdf / .pptx / .html 以及 .xlsx / .csv / .tsv 编码对照表正文，
|
||
按 kb.chunk_text 切块（跨块自动保留 Markdown 表头）并存入内容寻址工件（entity_doc_attachment）；
|
||
若勾选 save_to_kb=True，同步创建知识库 Document 并在后台完成切块向量索引，返回 document_id 供后续直接复用。
|
||
2. 已有知识库文档（Document + Chunk）：若文档仍在后台切块中，自动回退按 doc.content 即时切块，无需等待向量计算。
|
||
"""
|
||
from __future__ import annotations
|
||
|
||
import asyncio
|
||
import logging
|
||
import re
|
||
from collections.abc import Sequence
|
||
from dataclasses import dataclass
|
||
from typing import Any
|
||
|
||
from fastapi import BackgroundTasks, UploadFile
|
||
from sqlalchemy import select
|
||
from sqlalchemy.ext.asyncio import AsyncSession
|
||
|
||
from app.core import artifact_store
|
||
from app.core.config import settings
|
||
from app.core.errors import explain, raw as raw_err
|
||
from app.db.base import SessionLocal
|
||
from app.db.models import Chunk, Document
|
||
from app.memory import kb
|
||
from app.memory.parsing import UnsupportedDocument, extract
|
||
|
||
logger = logging.getLogger(__name__)
|
||
|
||
ATTACHMENT_KIND = "entity_doc_attachment"
|
||
MAX_DOC_SEARCH_LIMIT = 15
|
||
|
||
#: 持有后台切块任务引用，防止事件循环提前回收
|
||
_KB_TASKS: set[asyncio.Task[Any]] = set()
|
||
|
||
_TOKEN_RE = re.compile(r"[A-Za-z0-9_]+|[\u4e00-\u9fff]{2,}")
|
||
|
||
|
||
class AttachmentTooLarge(ValueError):
|
||
"""上传附件超过系统大小上限。"""
|
||
|
||
|
||
class AttachmentInvalid(ValueError):
|
||
"""上传附件格式不支持或无有效文本。"""
|
||
|
||
|
||
@dataclass(slots=True)
|
||
class EntityDocBundle:
|
||
ref_id: str
|
||
kind: str # "attachment" | "kb"
|
||
title: str
|
||
mime: str
|
||
char_count: int
|
||
chunks: list[str]
|
||
document_id: str | None = None
|
||
|
||
|
||
async def _index_kb_document(doc_id: str) -> None:
|
||
"""后台对新建的知识库文档切块并建立向量与倒排索引。"""
|
||
async with SessionLocal() as session:
|
||
doc = await session.get(Document, doc_id)
|
||
if not doc:
|
||
return
|
||
try:
|
||
await kb.process_document(session, doc)
|
||
except Exception as exc: # noqa: BLE001
|
||
logger.warning("实体文档存入知识库索引失败（%s）：%s", doc_id, raw_err(exc))
|
||
await session.rollback()
|
||
doc = await session.get(Document, doc_id)
|
||
if doc:
|
||
reason, hint = explain(exc)
|
||
doc.status = "failed"
|
||
doc.error = (f"文档分块或生成向量时出错：{reason}" + (f"。{hint}" if hint else ""))[:500]
|
||
await session.commit()
|
||
|
||
|
||
async def read_upload_limited(file: UploadFile) -> bytes:
|
||
limit = settings.max_upload_mb * 1024 * 1024
|
||
pieces: list[bytes] = []
|
||
total = 0
|
||
while chunk := await file.read(1024 * 1024):
|
||
total += len(chunk)
|
||
if total > limit:
|
||
raise AttachmentTooLarge(f"文件超过大小上限（{settings.max_upload_mb} MB）")
|
||
pieces.append(chunk)
|
||
return b"".join(pieces)
|
||
|
||
|
||
async def ingest_entity_attachment(
|
||
session: AsyncSession,
|
||
*,
|
||
raw_bytes: bytes,
|
||
filename: str,
|
||
mime: str = "",
|
||
save_to_kb: bool = False,
|
||
collection: str = "default",
|
||
background: BackgroundTasks | None = None,
|
||
) -> dict[str, Any]:
|
||
"""解析上传的业务文档或主数据表格，生成附件工件并按需存入知识库。"""
|
||
clean_name = (filename or "上传文档").strip() or "上传文档"
|
||
clean_mime = (mime or "text/plain").strip() or "text/plain"
|
||
if not raw_bytes:
|
||
raise AttachmentInvalid("上传文件为空，请选择包含有效内容的文件")
|
||
try:
|
||
content = extract(raw_bytes, clean_name, clean_mime, tables=True)
|
||
except UnsupportedDocument as exc:
|
||
raise AttachmentInvalid(str(exc)) from None
|
||
|
||
content = (content or "").strip()
|
||
if not content:
|
||
raise AttachmentInvalid("未能从文件中提取出有效文字或表格内容")
|
||
|
||
chunks = kb.chunk_text(content)
|
||
if not chunks:
|
||
chunks = [content[:4000]]
|
||
|
||
doc_id: str | None = None
|
||
clean_collection = (collection or "default").strip() or "default"
|
||
if save_to_kb:
|
||
doc = await kb.create_document(
|
||
session,
|
||
collection=clean_collection,
|
||
title=clean_name,
|
||
content=content,
|
||
source=clean_name,
|
||
mime=clean_mime,
|
||
meta={"origin": "entity_agent"},
|
||
status="processing",
|
||
)
|
||
doc_id = doc.id
|
||
if background is not None:
|
||
background.add_task(_index_kb_document, doc.id)
|
||
else:
|
||
task = asyncio.create_task(_index_kb_document(doc.id))
|
||
_KB_TASKS.add(task)
|
||
task.add_done_callback(_KB_TASKS.discard)
|
||
|
||
payload = {
|
||
"title": clean_name,
|
||
"filename": clean_name,
|
||
"mime": clean_mime,
|
||
"char_count": len(content),
|
||
"chunks": chunks,
|
||
"document_id": doc_id,
|
||
}
|
||
attachment_id = await artifact_store.put_json(
|
||
payload,
|
||
kind=ATTACHMENT_KIND,
|
||
meta={"title": clean_name, "document_id": doc_id},
|
||
)
|
||
return {
|
||
"attachment_id": attachment_id,
|
||
"title": clean_name,
|
||
"mime": clean_mime,
|
||
"char_count": len(content),
|
||
"chunk_count": len(chunks),
|
||
"saved_to_kb": bool(doc_id),
|
||
"document_id": doc_id,
|
||
"collection": clean_collection if doc_id else None,
|
||
}
|
||
|
||
|
||
async def load_doc_bundles(
|
||
session: AsyncSession,
|
||
*,
|
||
attachment_ids: Sequence[str] = (),
|
||
document_ids: Sequence[str] = (),
|
||
) -> list[EntityDocBundle]:
|
||
"""统一加载临时附件与知识库文档的切块，按引用去重。"""
|
||
bundles: list[EntityDocBundle] = []
|
||
seen_doc_ids: set[str] = set()
|
||
seen_att_ids: set[str] = set()
|
||
|
||
for raw_att in attachment_ids or ():
|
||
att_id = str(raw_att or "").strip()
|
||
if not att_id or att_id in seen_att_ids:
|
||
continue
|
||
seen_att_ids.add(att_id)
|
||
try:
|
||
loaded = artifact_store.load(att_id)
|
||
except ValueError:
|
||
loaded = None
|
||
if not isinstance(loaded, dict) or not isinstance(loaded.get("chunks"), list):
|
||
continue
|
||
chunks = [str(c) for c in loaded["chunks"] if isinstance(c, str) and c.strip()]
|
||
if not chunks:
|
||
continue
|
||
linked_doc_id = str(loaded.get("document_id") or "").strip() or None
|
||
if linked_doc_id:
|
||
seen_doc_ids.add(linked_doc_id)
|
||
bundles.append(
|
||
EntityDocBundle(
|
||
ref_id=att_id,
|
||
kind="attachment",
|
||
title=str(loaded.get("title") or "上传附件"),
|
||
mime=str(loaded.get("mime") or "text/plain"),
|
||
char_count=int(loaded.get("char_count") or sum(len(c) for c in chunks)),
|
||
chunks=chunks,
|
||
document_id=linked_doc_id,
|
||
)
|
||
)
|
||
|
||
clean_doc_ids = [
|
||
str(d).strip()
|
||
for d in (document_ids or ())
|
||
if str(d or "").strip() and str(d).strip() not in seen_doc_ids
|
||
]
|
||
if clean_doc_ids:
|
||
seen_doc_ids.update(clean_doc_ids)
|
||
docs = {
|
||
row.id: row
|
||
for row in (await session.execute(select(Document).where(Document.id.in_(clean_doc_ids)))).scalars()
|
||
}
|
||
chunk_rows = list(
|
||
(
|
||
await session.execute(
|
||
select(Chunk)
|
||
.where(Chunk.document_id.in_(clean_doc_ids))
|
||
.order_by(Chunk.document_id.asc(), Chunk.ordinal.asc())
|
||
)
|
||
).scalars()
|
||
)
|
||
by_doc: dict[str, list[str]] = {}
|
||
for crow in chunk_rows:
|
||
if crow.content and crow.content.strip():
|
||
by_doc.setdefault(crow.document_id, []).append(crow.content)
|
||
|
||
for doc_id in clean_doc_ids:
|
||
doc = docs.get(doc_id)
|
||
if doc is None:
|
||
continue
|
||
pieces = by_doc.get(doc_id) or kb.chunk_text(doc.content or "")
|
||
if not pieces and (doc.content or "").strip():
|
||
pieces = [(doc.content or "").strip()[:4000]]
|
||
if not pieces:
|
||
continue
|
||
bundles.append(
|
||
EntityDocBundle(
|
||
ref_id=doc.id,
|
||
kind="kb",
|
||
title=doc.title or doc.source or "知识库文档",
|
||
mime=doc.mime or "text/plain",
|
||
char_count=len(doc.content or ""),
|
||
chunks=pieces,
|
||
document_id=doc.id,
|
||
)
|
||
)
|
||
return bundles
|
||
|
||
|
||
def _keywords(query: str) -> list[str]:
|
||
q = (query or "").strip().lower()
|
||
if not q:
|
||
return []
|
||
tokens = [m.group(0) for m in _TOKEN_RE.finditer(q)]
|
||
out: list[str] = []
|
||
seen: set[str] = set()
|
||
for tok in [q, *tokens]:
|
||
if tok and tok not in seen:
|
||
seen.add(tok)
|
||
out.append(tok)
|
||
return out
|
||
|
||
|
||
def search_doc_bundles(
|
||
bundles: Sequence[EntityDocBundle],
|
||
*,
|
||
query: str = "",
|
||
doc_ref: str = "",
|
||
start_chunk: int = 0,
|
||
limit: int = 6,
|
||
) -> list[dict[str, Any]]:
|
||
"""按关键词或分段序号检索已挂载文档片段，返回带文档标题与段号的原文片段。"""
|
||
cap = max(1, min(int(limit or 6), MAX_DOC_SEARCH_LIMIT))
|
||
offset = max(0, int(start_chunk or 0))
|
||
ref_norm = (doc_ref or "").strip().lower()
|
||
|
||
scoped = [
|
||
b
|
||
for b in bundles
|
||
if not ref_norm
|
||
or b.ref_id.lower() == ref_norm
|
||
or (b.document_id and b.document_id.lower() == ref_norm)
|
||
or ref_norm in b.title.lower()
|
||
]
|
||
if not scoped:
|
||
scoped = list(bundles)
|
||
|
||
terms = _keywords(query)
|
||
if not terms:
|
||
out: list[dict[str, Any]] = []
|
||
for bundle in scoped:
|
||
for idx in range(offset, len(bundle.chunks)):
|
||
out.append({
|
||
"doc_ref": bundle.ref_id,
|
||
"title": bundle.title,
|
||
"chunk_index": idx,
|
||
"total_chunks": len(bundle.chunks),
|
||
"citation": f"文档《{bundle.title}》第 {idx + 1} 段",
|
||
"content": bundle.chunks[idx],
|
||
})
|
||
if len(out) >= cap:
|
||
return out
|
||
return out
|
||
|
||
scored: list[tuple[int, int, int, EntityDocBundle, str]] = []
|
||
for b_idx, bundle in enumerate(scoped):
|
||
for c_idx, chunk in enumerate(bundle.chunks):
|
||
if c_idx < offset:
|
||
continue
|
||
low = chunk.lower()
|
||
score = 0
|
||
for t_idx, term in enumerate(terms):
|
||
hits = low.count(term)
|
||
if hits:
|
||
weight = 5 if t_idx == 0 else (3 if len(term) >= 3 else 1)
|
||
score += hits * weight
|
||
if score > 0:
|
||
scored.append((-score, b_idx, c_idx, bundle, chunk))
|
||
scored.sort(key=lambda item: (item[0], item[1], item[2]))
|
||
return [
|
||
{
|
||
"doc_ref": bundle.ref_id,
|
||
"title": bundle.title,
|
||
"chunk_index": c_idx,
|
||
"total_chunks": len(bundle.chunks),
|
||
"citation": f"文档《{bundle.title}》第 {c_idx + 1} 段",
|
||
"content": chunk,
|
||
}
|
||
for _neg, _b, c_idx, bundle, chunk in scored[:cap]
|
||
]
|