v1.0.155: VM PostgreSQL + modular db/ & services/ — Lucee stays SPA shell only
This commit is contained in:
+75
-493
@@ -1,59 +1,31 @@
|
||||
#!/usr/bin/env python3
|
||||
"""HTTP-сервер: конвертация .doc → .docx + LLM-анализ ДС (/llm-ops) + SSE v2 (/process-v2)"""
|
||||
"""Contracts VM server — thin HTTP router. All logic in db/ and services/."""
|
||||
from http.server import HTTPServer, BaseHTTPRequestHandler
|
||||
from socketserver import ThreadingMixIn
|
||||
|
||||
class ThreadingHTTPServer(ThreadingMixIn, HTTPServer):
|
||||
"""Многопоточный HTTP-сервер."""
|
||||
daemon_threads = True
|
||||
from urllib.parse import urlparse, parse_qs
|
||||
import subprocess, tempfile, os, sys, json, time, re
|
||||
import json, re
|
||||
|
||||
import httpx
|
||||
# ── DB auto-seed ──────────────────────────────────────────────────────────
|
||||
from db import prompts as db_prompts
|
||||
from db.connection import DB_CONFIG
|
||||
db_prompts.seed_defaults()
|
||||
|
||||
# ── Services ──────────────────────────────────────────────────────────────
|
||||
from services.upload import handle_upload
|
||||
from services.unzip import handle_unzip
|
||||
from services.process import run_pipeline
|
||||
from llm_prompt import build_prompt
|
||||
|
||||
LLM_URL = "https://api.aillm.ru/v1/chat/completions"
|
||||
LLM_KEY = os.environ.get("LLM_KEY", "")
|
||||
LLM_MODEL = "gpt-oss-120b"
|
||||
LUCEE_URL = "https://contractor.luceek8s.dev.nubes.ru"
|
||||
|
||||
|
||||
def call_llm(current_spec, doc_text):
|
||||
"""Вызов LLM API. Возвращает ({mode, ops}, prompt_id) или кидает исключение."""
|
||||
prompt, prompt_id = build_prompt(current_spec, doc_text)
|
||||
payload = {
|
||||
"model": LLM_MODEL,
|
||||
"messages": [{"role": "user", "content": prompt}],
|
||||
"max_tokens": 8000,
|
||||
"temperature": 0.1,
|
||||
}
|
||||
with httpx.Client(http2=True, timeout=120, verify=False) as client:
|
||||
resp = client.post(LLM_URL, json=payload,
|
||||
headers={"Authorization": f"Bearer {LLM_KEY}", "Content-Type": "application/json"})
|
||||
resp.raise_for_status()
|
||||
data = resp.json()
|
||||
|
||||
raw_text = data.get("choices", [{}])[0].get("message", {}).get("content", "")
|
||||
json_text = raw_text
|
||||
if "```json" in json_text:
|
||||
json_text = json_text.split("```json")[1].split("```")[0]
|
||||
elif "```" in json_text:
|
||||
json_text = json_text.split("```")[1].split("```")[0]
|
||||
return json.loads(json_text.strip()), prompt_id
|
||||
class ThreadingHTTPServer(ThreadingMixIn, HTTPServer):
|
||||
daemon_threads = True
|
||||
|
||||
|
||||
class Handler(BaseHTTPRequestHandler):
|
||||
def do_GET(self):
|
||||
parsed = urlparse(self.path)
|
||||
if parsed.path == "/process-v2":
|
||||
self._handle_process_v2(parsed)
|
||||
else:
|
||||
self.send_error(404)
|
||||
|
||||
def do_OPTIONS(self):
|
||||
self.send_response(200)
|
||||
self.send_header("Access-Control-Allow-Origin", "*")
|
||||
self.send_header("Access-Control-Allow-Methods", "GET, OPTIONS")
|
||||
self._send_cors()
|
||||
self.send_header("Access-Control-Allow-Methods", "GET, POST, OPTIONS")
|
||||
self.send_header("Access-Control-Allow-Headers", "Content-Type")
|
||||
self.end_headers()
|
||||
|
||||
@@ -61,50 +33,29 @@ class Handler(BaseHTTPRequestHandler):
|
||||
parsed = urlparse(self.path)
|
||||
if parsed.path == "/process-v2":
|
||||
self._handle_process_v2(parsed)
|
||||
elif parsed.path == "/health":
|
||||
self._json({"ok": True, "db": DB_CONFIG["dbname"]})
|
||||
else:
|
||||
self.send_error(404)
|
||||
|
||||
def do_POST(self):
|
||||
if self.path == "/llm-ops":
|
||||
self._handle_llm_ops()
|
||||
elif self.path == "/unzip-upload":
|
||||
self._handle_unzip_upload()
|
||||
elif self.path == "/upload":
|
||||
if self.path == "/upload":
|
||||
self._handle_upload()
|
||||
elif self.path == "/parse-pdf":
|
||||
self._handle_parse_pdf()
|
||||
elif self.path == "/unzip-upload":
|
||||
self._handle_unzip()
|
||||
elif self.path == "/llm-ops":
|
||||
self._handle_llm_ops()
|
||||
else:
|
||||
self._handle_doc_convert()
|
||||
self.send_error(404)
|
||||
|
||||
def _sse(self, data):
|
||||
"""Отправить SSE-событие."""
|
||||
self.wfile.write(f"data: {json.dumps(data, ensure_ascii=False)}\n\n".encode())
|
||||
self.wfile.flush()
|
||||
|
||||
def _lucee_query(self, sql):
|
||||
"""Запрос к Lucee API."""
|
||||
resp = httpx.get(f"{LUCEE_URL}/api.cfm", params={"action": "query", "sql": sql}, timeout=30)
|
||||
resp.raise_for_status()
|
||||
d = resp.json()
|
||||
if not d.get("OK"):
|
||||
raise Exception(d.get("ERROR", "API error"))
|
||||
cols = d["COLUMNS"]
|
||||
rows = []
|
||||
for r in d["ROWS"]:
|
||||
row = {}
|
||||
for i, c in enumerate(cols):
|
||||
row[c.lower()] = r[i] if isinstance(r, list) else r[c]
|
||||
rows.append(row)
|
||||
return rows
|
||||
# ── /process-v2 (SSE) ─────────────────────────────────────────────────
|
||||
|
||||
def _handle_process_v2(self, parsed):
|
||||
"""SSE-стриминг v2 пайплайна."""
|
||||
params = parse_qs(parsed.query)
|
||||
cid = params.get("contract_id", [None])[0]
|
||||
if not cid:
|
||||
self.send_error(400, "contract_id required")
|
||||
return
|
||||
# Validate UUID to prevent SQL injection
|
||||
if not re.fullmatch(r'[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}', cid, re.I):
|
||||
self.send_error(400, "invalid contract_id format")
|
||||
return
|
||||
@@ -117,445 +68,76 @@ class Handler(BaseHTTPRequestHandler):
|
||||
self.wfile.write(b": ok\n\n")
|
||||
self.wfile.flush()
|
||||
|
||||
t0 = time.time()
|
||||
order_ids = params.get("order", [None])[0] or ""
|
||||
|
||||
try:
|
||||
# 0. Сброс предыдущих результатов сравнения
|
||||
r = httpx.get(f"{LUCEE_URL}/reset_contract.cfm?contract_id={cid}", timeout=10)
|
||||
if r.status_code != 200:
|
||||
self._sse({"type": "error", "message": f"reset failed: HTTP {r.status_code}"})
|
||||
return
|
||||
|
||||
# 1. Список supplements (с document_id для provenance)
|
||||
supps = self._lucee_query(
|
||||
f"SELECT s.id, s.type, d.filename, d.id as document_id FROM supplements s "
|
||||
f"JOIN documents d ON s.document_id=d.id "
|
||||
f"WHERE s.contract_id='{cid}' AND d.elements_json IS NOT NULL ORDER BY s.created_at"
|
||||
)
|
||||
# Reorder if client specified &order=...
|
||||
order_ids = (params.get("order", [None])[0] or "").strip()
|
||||
if order_ids:
|
||||
order_list = [x.strip() for x in order_ids.split(",") if x.strip()]
|
||||
order_map = {oid: i for i, oid in enumerate(order_list)}
|
||||
supps.sort(key=lambda s: order_map.get(s["id"], 999999))
|
||||
if not supps:
|
||||
self._sse({"type": "error", "message": "Нет распарсенных файлов"})
|
||||
return
|
||||
|
||||
for s in supps:
|
||||
sid = s["id"]
|
||||
|
||||
# Текущая спецификация
|
||||
cur = self._lucee_query(
|
||||
f"SELECT name_hash, name, price, qty, sum, date_start "
|
||||
f"FROM spec_current WHERE contract_id='{cid}' ORDER BY name"
|
||||
)
|
||||
current_spec = []
|
||||
for r in cur:
|
||||
current_spec.append({
|
||||
"hash": r["name_hash"],
|
||||
"name": r["name"],
|
||||
"price": float(r["price"]) if r.get("price") else None,
|
||||
"qty": float(r["qty"]) if r.get("qty") else None,
|
||||
"sum": float(r["sum"]) if r.get("sum") else None,
|
||||
"date_start": r["date_start"],
|
||||
})
|
||||
|
||||
# elements_json → текст
|
||||
docs = self._lucee_query(
|
||||
f"SELECT elements_json FROM documents d "
|
||||
f"JOIN supplements s ON s.document_id=d.id WHERE s.id='{sid}'"
|
||||
)
|
||||
if not docs:
|
||||
continue
|
||||
ej = docs[0]["elements_json"]
|
||||
# Lucee JSONB может вернуть: строку, dict-обёртку {Value,Type}, или уже список
|
||||
if isinstance(ej, dict) and "Value" in ej:
|
||||
ej = ej["Value"]
|
||||
if isinstance(ej, str):
|
||||
elements = json.loads(ej)
|
||||
elif isinstance(ej, list):
|
||||
elements = ej
|
||||
else:
|
||||
elements = []
|
||||
# Lucee serializeJSON → UPPERCASE keys, normalize to lowercase
|
||||
elements = [{k.lower(): v for k, v in el.items()} for el in elements]
|
||||
lines = []
|
||||
for el in elements:
|
||||
if el.get("type") == "paragraph":
|
||||
prefix = f"[{el['style']}] " if el.get("style") else ""
|
||||
lines.append(prefix + el.get("text", ""))
|
||||
elif el.get("type") == "table":
|
||||
rows = el.get("rows", [])
|
||||
if rows:
|
||||
ncols = len(rows[0])
|
||||
lines.append(f"--- Таблица ({len(rows)}×{ncols}) ---")
|
||||
for row in rows:
|
||||
cells = [str(c).replace("\n", " ").replace("|", "\\|") for c in row[:ncols]]
|
||||
lines.append("| " + " | ".join(cells) + " |")
|
||||
lines.append("")
|
||||
doc_text = "\n".join(lines)
|
||||
|
||||
self._sse({"type": "extract_start", "supplement_id": sid, "filename": s["filename"]})
|
||||
|
||||
# LLM (прямой вызов, не HTTP self-call)
|
||||
import threading
|
||||
t1 = time.time()
|
||||
done = threading.Event()
|
||||
result = [None]
|
||||
error = [None]
|
||||
|
||||
def do_llm():
|
||||
try:
|
||||
result[0] = call_llm(current_spec, doc_text)
|
||||
except Exception as e:
|
||||
error[0] = str(e)
|
||||
finally:
|
||||
done.set()
|
||||
|
||||
t = threading.Thread(target=do_llm)
|
||||
t.start()
|
||||
|
||||
# Keepalive каждые 15с пока LLM думает
|
||||
while not done.wait(15):
|
||||
self._sse({"type": "keepalive"})
|
||||
|
||||
t.join()
|
||||
elapsed = round(time.time() - t1, 1)
|
||||
|
||||
if error[0]:
|
||||
self._sse({"type": "extract_error", "supplement_id": sid, "filename": s["filename"],
|
||||
"error": error[0], "time_s": elapsed})
|
||||
continue
|
||||
|
||||
llm_result, prompt_id = result[0]
|
||||
ops = llm_result.get("ops", [])
|
||||
mode = llm_result.get("mode", "partial")
|
||||
|
||||
# Маппинг target_id (r1, r2...) → target_hash из current_spec
|
||||
id_map = {f"r{i+1}": r["hash"] for i, r in enumerate(current_spec)}
|
||||
# Обогащение ops: target_id → target_hash + имя из current_spec
|
||||
spec_names = {r["hash"]: r["name"] for r in current_spec}
|
||||
for op in ops:
|
||||
tid = op.get("target_id")
|
||||
if tid and tid in id_map:
|
||||
op["target_hash"] = id_map[tid]
|
||||
# Добавить имя для UPDATE/DELETE из spec_current
|
||||
th = op.get("target_hash")
|
||||
if th and th in spec_names and not op.get("new_row", {}).get("name"):
|
||||
op.setdefault("new_row", {})["name"] = spec_names[th]
|
||||
self._sse({"type": "llm_done", "supplement_id": sid, "filename": s["filename"],
|
||||
"ops_count": len(ops), "mode": mode, "time_s": elapsed})
|
||||
|
||||
# Применить (с provenance: document_id, prompt_id, raw_llm_response)
|
||||
apply_resp = httpx.post(
|
||||
f"{LUCEE_URL}/apply_events.cfm?contract_id={cid}&mode={mode}",
|
||||
json={
|
||||
"supplement_id": sid,
|
||||
"document_id": s.get("document_id", ""),
|
||||
"ops": ops,
|
||||
"prompt_id": prompt_id,
|
||||
"raw_llm_response": llm_result
|
||||
}, timeout=30
|
||||
)
|
||||
if apply_resp.status_code == 200:
|
||||
ar = apply_resp.json()
|
||||
# Lucee serializeJSON → UPPERCASE keys
|
||||
summary = {k.lower(): v for k, v in ar.get("SUMMARY", {}).items()}
|
||||
self._sse({"type": "applied", "supplement_id": sid, "summary": summary, "ops": ops})
|
||||
else:
|
||||
self._sse({"type": "apply_error", "supplement_id": sid,
|
||||
"error": f"apply HTTP {apply_resp.status_code}"})
|
||||
|
||||
total_time = round(time.time() - t0, 1)
|
||||
self._sse({"type": "done", "total_time_s": total_time})
|
||||
|
||||
run_pipeline(cid, order_ids, self._sse, build_prompt)
|
||||
except Exception as e:
|
||||
self._sse({"type": "error", "message": str(e)})
|
||||
|
||||
def _sse(self, data):
|
||||
self.wfile.write(f"data: {json.dumps(data, ensure_ascii=False)}\n\n".encode())
|
||||
self.wfile.flush()
|
||||
|
||||
# ── /upload ───────────────────────────────────────────────────────────
|
||||
|
||||
def _handle_upload(self):
|
||||
try:
|
||||
content_type = self.headers.get("Content-Type", "")
|
||||
content_length = int(self.headers.get("Content-Length", 0))
|
||||
result = handle_upload(self.rfile, content_type, content_length)
|
||||
self._json(result)
|
||||
except Exception as e:
|
||||
self._json({"ok": False, "error": str(e)}, 500)
|
||||
|
||||
# ── /unzip-upload ─────────────────────────────────────────────────────
|
||||
|
||||
def _handle_unzip(self):
|
||||
try:
|
||||
content_length = int(self.headers.get("Content-Length", 0))
|
||||
result = handle_unzip(self.rfile, content_length)
|
||||
self._json(result)
|
||||
except Exception as e:
|
||||
self._json({"ok": False, "error": str(e)}, 500)
|
||||
|
||||
# ── /llm-ops (debug) ──────────────────────────────────────────────────
|
||||
|
||||
def _handle_llm_ops(self):
|
||||
from services.llm import call_llm
|
||||
length = int(self.headers.get("Content-Length", 0))
|
||||
body = json.loads(self.rfile.read(length))
|
||||
try:
|
||||
result, prompt_id = call_llm(body.get("current_spec", []), body.get("doc_text", ""))
|
||||
self.send_response(200)
|
||||
self.send_header("Content-Type", "application/json")
|
||||
self.end_headers()
|
||||
self.wfile.write(json.dumps(result, ensure_ascii=False).encode())
|
||||
except Exception as e:
|
||||
self.send_response(502)
|
||||
self.send_header("Content-Type", "application/json")
|
||||
self.end_headers()
|
||||
self.wfile.write(json.dumps({"error": str(e)}, ensure_ascii=False).encode())
|
||||
|
||||
def _handle_upload(self):
|
||||
"""Принять файл и передать в Lucee чанками (лимит ingress ~30KB)."""
|
||||
import uuid, base64
|
||||
from email.parser import BytesParser
|
||||
|
||||
content_type = self.headers.get("Content-Type", "")
|
||||
length = int(self.headers.get("Content-Length", 0))
|
||||
body = self.rfile.read(length)
|
||||
|
||||
msg = BytesParser().parsebytes(
|
||||
b"Content-Type: " + content_type.encode() + b"\r\n\r\n" + body
|
||||
)
|
||||
filename = None
|
||||
file_data = None
|
||||
contract_id = ""
|
||||
if msg.is_multipart():
|
||||
for part in msg.get_payload():
|
||||
fn = part.get_filename()
|
||||
if fn:
|
||||
filename = fn
|
||||
file_data = part.get_payload(decode=True)
|
||||
else:
|
||||
disp = part.get("Content-Disposition", "")
|
||||
if "name=\"contract_id\"" in disp:
|
||||
contract_id = part.get_payload(decode=True).decode("utf-8", errors="ignore").strip()
|
||||
|
||||
if not filename or not file_data:
|
||||
self._send_error_cors(400, "no file in request")
|
||||
return
|
||||
|
||||
CHUNK = 10000 # 10KB raw → 20KB hex → помещается
|
||||
upload_id = str(uuid.uuid4())
|
||||
raw = file_data
|
||||
total = (len(raw) + CHUNK - 1) // CHUNK
|
||||
|
||||
for i in range(total):
|
||||
chunk = raw[i*CHUNK:(i+1)*CHUNK]
|
||||
with httpx.Client(http2=True, timeout=30, verify=False) as client:
|
||||
r = client.post(
|
||||
f"{LUCEE_URL}/chunk.cfm?action=put",
|
||||
json={
|
||||
"upload_id": upload_id,
|
||||
"chunk_index": i,
|
||||
"total_chunks": total,
|
||||
"filename": filename,
|
||||
"data": base64.b64encode(chunk).decode(),
|
||||
},
|
||||
)
|
||||
if r.status_code != 200:
|
||||
self._send_error_cors(502, f"chunk {i+1}/{total} failed: HTTP {r.status_code}")
|
||||
return
|
||||
|
||||
with httpx.Client(http2=True, timeout=60, verify=False) as client:
|
||||
resp = client.post(
|
||||
f"{LUCEE_URL}/chunk.cfm?action=assemble",
|
||||
json={"upload_id": upload_id, "contract_id": contract_id},
|
||||
result, prompt_id = call_llm(
|
||||
body.get("current_spec", []),
|
||||
body.get("doc_text", ""),
|
||||
build_prompt,
|
||||
)
|
||||
if resp.status_code == 200:
|
||||
result = resp.json()
|
||||
doc_id = result.get("DOC_ID", "")
|
||||
|
||||
# Парсинг на VM + сохранить в Lucee
|
||||
parsed = self._parse_file(filename, file_data)
|
||||
if parsed and parsed.get("status") == "parsed" and doc_id:
|
||||
try:
|
||||
escaped = json.dumps(parsed["elements"], ensure_ascii=False).replace("'", "''")
|
||||
with httpx.Client(http2=True, timeout=15, verify=False) as c:
|
||||
c.post(
|
||||
f"{LUCEE_URL}/api.cfm",
|
||||
data={
|
||||
"action": "execute",
|
||||
"sql": f"UPDATE documents SET elements_json='{escaped}'::jsonb, status='parsed' WHERE id='{doc_id}'"
|
||||
}
|
||||
)
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
result["parsed"] = parsed
|
||||
self.send_response(200)
|
||||
self._send_cors()
|
||||
self.send_header("Content-Type", "application/json; charset=utf-8")
|
||||
self.end_headers()
|
||||
self.wfile.write(json.dumps(result, ensure_ascii=False).encode())
|
||||
else:
|
||||
self._send_error_cors(502, f"assemble failed: HTTP {resp.status_code}")
|
||||
|
||||
def _parse_file(self, filename, data):
|
||||
"""Распарсить файл на VM: PDF→PyPDF2, DOCX/DOC→python-docx, иначе plain text."""
|
||||
import base64
|
||||
ext = filename.rsplit('.', 1)[-1].lower() if '.' in filename else ''
|
||||
try:
|
||||
if ext == 'pdf':
|
||||
from PyPDF2 import PdfReader
|
||||
import io as io_mod
|
||||
reader = PdfReader(io_mod.BytesIO(data))
|
||||
elements = []
|
||||
for page in reader.pages:
|
||||
text = page.extract_text()
|
||||
if text:
|
||||
for line in text.split('\n'):
|
||||
line = line.strip()
|
||||
if line:
|
||||
elements.append({"type": "paragraph", "text": line, "style": ""})
|
||||
return {"status": "parsed", "element_count": len(elements), "elements": elements}
|
||||
elif ext in ('docx', 'doc'):
|
||||
# Для DOCX/DOC — оставляем Lucee (Apache POI работает)
|
||||
return None
|
||||
else:
|
||||
# Plain text
|
||||
text = data.decode('utf-8', errors='ignore')[:5000]
|
||||
lines = [l.strip() for l in text.split('\n') if l.strip()]
|
||||
return {"status": "parsed", "element_count": len(lines),
|
||||
"elements": [{"type": "paragraph", "text": l, "style": ""} for l in lines]}
|
||||
self._json(result)
|
||||
except Exception as e:
|
||||
return {"status": "error", "error": str(e), "element_count": 0}
|
||||
self._json({"error": str(e)}, 502)
|
||||
|
||||
def _handle_parse_pdf(self):
|
||||
"""Распарсить PDF через PyPDF2 (получает base64 тело от Lucee)."""
|
||||
import base64
|
||||
from PyPDF2 import PdfReader
|
||||
import io as io_module
|
||||
# ── Helpers ───────────────────────────────────────────────────────────
|
||||
|
||||
length = int(self.headers.get("Content-Length", 0))
|
||||
body = self.rfile.read(length).decode("utf-8", errors="ignore").strip()
|
||||
raw = base64.b64decode(body)
|
||||
|
||||
try:
|
||||
reader = PdfReader(io_module.BytesIO(raw))
|
||||
elements = []
|
||||
for page in reader.pages:
|
||||
text = page.extract_text()
|
||||
if text:
|
||||
for line in text.split('\n'):
|
||||
line = line.strip()
|
||||
if line:
|
||||
elements.append({"type": "paragraph", "text": line, "style": ""})
|
||||
result = {
|
||||
"ok": True,
|
||||
"status": "parsed",
|
||||
"element_count": len(elements),
|
||||
"elements": elements,
|
||||
"tables": 0,
|
||||
"paragraphs": len(elements),
|
||||
}
|
||||
except Exception as e:
|
||||
result = {"ok": False, "status": "error", "error": str(e), "element_count": 0}
|
||||
|
||||
self.send_response(200)
|
||||
self.send_header("Content-Type", "application/json; charset=utf-8")
|
||||
self.send_header("Access-Control-Allow-Origin", "*")
|
||||
self.end_headers()
|
||||
self.wfile.write(json.dumps(result, ensure_ascii=False).encode())
|
||||
|
||||
def _handle_unzip_upload(self):
|
||||
"""Распаковать ZIP и загрузить каждый файл в Lucee. Только файлы из корня архива."""
|
||||
import zipfile, io
|
||||
from email.parser import BytesParser
|
||||
|
||||
content_type = self.headers.get("Content-Type", "")
|
||||
length = int(self.headers.get("Content-Length", 0))
|
||||
body = self.rfile.read(length)
|
||||
|
||||
# Raw binary (браузер) или multipart (curl fallback)
|
||||
if "multipart" in content_type:
|
||||
msg = BytesParser().parsebytes(
|
||||
b"Content-Type: " + content_type.encode() + b"\r\n\r\n" + body
|
||||
)
|
||||
zip_data = None
|
||||
if msg.is_multipart():
|
||||
for part in msg.get_payload():
|
||||
if part.get_filename():
|
||||
zip_data = part.get_payload(decode=True)
|
||||
break
|
||||
else:
|
||||
zip_data = body
|
||||
else:
|
||||
zip_data = body
|
||||
|
||||
if not zip_data:
|
||||
self._send_error_cors(400, "no file in request")
|
||||
return
|
||||
|
||||
results = []
|
||||
try:
|
||||
with zipfile.ZipFile(io.BytesIO(zip_data)) as zf:
|
||||
for name in zf.namelist():
|
||||
if name.endswith('/') or '/' in name:
|
||||
continue
|
||||
content = zf.read(name)
|
||||
ext = name.rsplit('.', 1)[-1].lower() if '.' in name else ''
|
||||
if ext not in ('docx', 'doc', 'pdf'):
|
||||
continue
|
||||
|
||||
import base64 as b64
|
||||
content_b64 = b64.b64encode(content).decode()
|
||||
resp = httpx.post(
|
||||
f"{LUCEE_URL}/upload.cfm",
|
||||
data={"filename": name, "data": content_b64},
|
||||
timeout=30
|
||||
)
|
||||
if resp.status_code == 200:
|
||||
rj = resp.json()
|
||||
if rj.get("OK"):
|
||||
entry = {
|
||||
"filename": name,
|
||||
"doc_id": rj.get("DOC_ID", ""),
|
||||
"contract_id": rj.get("CONTRACT_ID", ""),
|
||||
"size": len(content),
|
||||
}
|
||||
# Парсинг PDF на VM
|
||||
parsed = self._parse_file(name, content)
|
||||
if parsed:
|
||||
entry["parsed"] = parsed
|
||||
results.append(entry)
|
||||
else:
|
||||
results.append({"filename": name, "error": rj.get("ERROR", "upload failed")})
|
||||
else:
|
||||
results.append({"filename": name, "error": f"HTTP {resp.status_code}"})
|
||||
|
||||
except zipfile.BadZipFile:
|
||||
self._send_error_cors(400, "not a valid ZIP file")
|
||||
return
|
||||
except Exception as e:
|
||||
self._send_error_cors(500, str(e))
|
||||
return
|
||||
|
||||
self.send_response(200)
|
||||
def _json(self, data, status=200):
|
||||
self.send_response(status)
|
||||
self._send_cors()
|
||||
self.send_header("Content-Type", "application/json; charset=utf-8")
|
||||
self.end_headers()
|
||||
self.wfile.write(json.dumps({"ok": True, "files": results}, ensure_ascii=False).encode())
|
||||
self.wfile.write(json.dumps(data, ensure_ascii=False).encode())
|
||||
|
||||
def _send_cors(self):
|
||||
self.send_header("Access-Control-Allow-Origin", "*")
|
||||
self.send_header("Access-Control-Allow-Methods", "POST, OPTIONS")
|
||||
self.send_header("Access-Control-Allow-Headers", "*")
|
||||
|
||||
def _send_error_cors(self, code, message):
|
||||
self.send_response(code)
|
||||
self._send_cors()
|
||||
self.send_header("Content-Type", "application/json; charset=utf-8")
|
||||
self.end_headers()
|
||||
self.wfile.write(json.dumps({"ok": False, "error": message}, ensure_ascii=False).encode())
|
||||
def log_message(self, format, *args):
|
||||
pass
|
||||
|
||||
def _handle_doc_convert(self):
|
||||
length = int(self.headers.get("Content-Length", 0))
|
||||
data = self.rfile.read(length)
|
||||
with tempfile.NamedTemporaryFile(suffix=".doc", delete=False) as f:
|
||||
f.write(data); doc_path = f.name
|
||||
tmpdir = tempfile.mkdtemp()
|
||||
try:
|
||||
subprocess.run(["libreoffice", "--headless", "--convert-to", "docx", "--outdir", tmpdir, doc_path],
|
||||
timeout=30, capture_output=True)
|
||||
docx = [x for x in os.listdir(tmpdir) if x.endswith(".docx")]
|
||||
if docx:
|
||||
with open(os.path.join(tmpdir, docx[0]), "rb") as f:
|
||||
out = f.read()
|
||||
self.send_response(200)
|
||||
self.send_header("Content-Type", "application/vnd.openxmlformats-officedocument.wordprocessingml.document")
|
||||
self.send_header("Content-Length", len(out))
|
||||
self.end_headers()
|
||||
self.wfile.write(out)
|
||||
else:
|
||||
self.send_error(500, "Conversion failed")
|
||||
finally:
|
||||
os.unlink(doc_path)
|
||||
for x in os.listdir(tmpdir): os.unlink(os.path.join(tmpdir, x))
|
||||
os.rmdir(tmpdir)
|
||||
def log_message(self, *a): pass
|
||||
|
||||
ThreadingHTTPServer(("127.0.0.1", 8766), Handler).serve_forever()
|
||||
if __name__ == "__main__":
|
||||
import os
|
||||
port = int(os.environ.get("PORT", "8766"))
|
||||
server = ThreadingHTTPServer(("0.0.0.0", port), Handler)
|
||||
print(f"Contracts VM server on :{port}, db={DB_CONFIG['dbname']}")
|
||||
try:
|
||||
server.serve_forever()
|
||||
except KeyboardInterrupt:
|
||||
server.shutdown()
|
||||
|
||||
@@ -0,0 +1,70 @@
|
||||
"""Database connection — psycopg2 connection pool."""
|
||||
import os
|
||||
import psycopg2
|
||||
import psycopg2.pool
|
||||
import psycopg2.extras
|
||||
|
||||
_pool = None
|
||||
|
||||
DB_CONFIG = {
|
||||
"host": os.getenv("DB_HOST", "127.0.0.1"),
|
||||
"port": int(os.getenv("DB_PORT", "5432")),
|
||||
"dbname": os.getenv("DB_NAME", "baza"),
|
||||
"user": os.getenv("DB_USER", "super"),
|
||||
"password": os.getenv("DB_PASS", ""),
|
||||
}
|
||||
|
||||
|
||||
def get_pool():
|
||||
global _pool
|
||||
if _pool is None:
|
||||
_pool = psycopg2.pool.ThreadedConnectionPool(
|
||||
minconn=1, maxconn=10, **DB_CONFIG
|
||||
)
|
||||
return _pool
|
||||
|
||||
|
||||
def query(sql, params=None):
|
||||
"""SELECT → list of dicts (lowercase keys)."""
|
||||
pool = get_pool()
|
||||
conn = pool.getconn()
|
||||
try:
|
||||
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
|
||||
cur.execute(sql, params)
|
||||
rows = cur.fetchall()
|
||||
return [{k.lower(): v for k, v in r.items()} for r in rows]
|
||||
finally:
|
||||
pool.putconn(conn)
|
||||
|
||||
|
||||
def execute(sql, params=None):
|
||||
"""INSERT/UPDATE/DELETE → rowcount."""
|
||||
pool = get_pool()
|
||||
conn = pool.getconn()
|
||||
try:
|
||||
with conn.cursor() as cur:
|
||||
cur.execute(sql, params)
|
||||
conn.commit()
|
||||
return cur.rowcount
|
||||
except Exception:
|
||||
conn.rollback()
|
||||
raise
|
||||
finally:
|
||||
pool.putconn(conn)
|
||||
|
||||
|
||||
def execute_returning(sql, params=None):
|
||||
"""INSERT/UPDATE/DELETE with RETURNING → first row dict."""
|
||||
pool = get_pool()
|
||||
conn = pool.getconn()
|
||||
try:
|
||||
with conn.cursor(cursor_factory=psycopg2.extras.RealDictCursor) as cur:
|
||||
cur.execute(sql, params)
|
||||
conn.commit()
|
||||
row = cur.fetchone()
|
||||
return {k.lower(): v for k, v in row.items()} if row else None
|
||||
except Exception:
|
||||
conn.rollback()
|
||||
raise
|
||||
finally:
|
||||
pool.putconn(conn)
|
||||
@@ -0,0 +1,25 @@
|
||||
"""Contracts CRUD."""
|
||||
from .connection import query, execute, execute_returning
|
||||
|
||||
|
||||
def insert(number, client=""):
|
||||
return execute_returning(
|
||||
"INSERT INTO contracts (number, client) VALUES (%s, %s) RETURNING *",
|
||||
(number, client),
|
||||
)
|
||||
|
||||
|
||||
def get(contract_id):
|
||||
rows = query("SELECT * FROM contracts WHERE id = %s", (contract_id,))
|
||||
return rows[0] if rows else None
|
||||
|
||||
|
||||
def delete(contract_id):
|
||||
return execute("DELETE FROM contracts WHERE id = %s", (contract_id,))
|
||||
|
||||
|
||||
def delete_orphaned():
|
||||
"""Remove contracts with no supplements."""
|
||||
execute(
|
||||
"DELETE FROM contracts WHERE id NOT IN (SELECT DISTINCT contract_id FROM supplements)"
|
||||
)
|
||||
@@ -0,0 +1,36 @@
|
||||
"""Documents CRUD."""
|
||||
from .connection import query, execute, execute_returning
|
||||
|
||||
|
||||
def insert(filename, mime_type, original_bytes, status="uploaded"):
|
||||
"""Insert document, return row dict."""
|
||||
return execute_returning(
|
||||
"""INSERT INTO documents (filename, mime_type, original_bytes, status)
|
||||
VALUES (%s, %s, %s, %s) RETURNING *""",
|
||||
(filename, mime_type, original_bytes, status),
|
||||
)
|
||||
|
||||
|
||||
def get(doc_id):
|
||||
rows = query("SELECT * FROM documents WHERE id = %s", (doc_id,))
|
||||
return rows[0] if rows else None
|
||||
|
||||
|
||||
def set_parsed(doc_id, elements_json):
|
||||
"""Update elements_json + status='parsed'."""
|
||||
import json
|
||||
return execute(
|
||||
"UPDATE documents SET elements_json = %s::jsonb, status = 'parsed' WHERE id = %s",
|
||||
(json.dumps(elements_json, ensure_ascii=False), doc_id),
|
||||
)
|
||||
|
||||
|
||||
def set_error(doc_id, error_message):
|
||||
return execute(
|
||||
"UPDATE documents SET status = 'error', error_message = %s WHERE id = %s",
|
||||
(error_message, doc_id),
|
||||
)
|
||||
|
||||
|
||||
def delete(doc_id):
|
||||
return execute("DELETE FROM documents WHERE id = %s", (doc_id,))
|
||||
@@ -0,0 +1,37 @@
|
||||
"""Prompts CRUD."""
|
||||
from .connection import query, execute, execute_returning
|
||||
|
||||
|
||||
def get_active(role):
|
||||
rows = query(
|
||||
"SELECT * FROM prompts WHERE role = %s AND is_active = true ORDER BY created_at DESC LIMIT 1",
|
||||
(role,),
|
||||
)
|
||||
return rows[0] if rows else None
|
||||
|
||||
|
||||
def get(prompt_id):
|
||||
rows = query("SELECT * FROM prompts WHERE id = %s", (prompt_id,))
|
||||
return rows[0] if rows else None
|
||||
|
||||
|
||||
def seed_defaults():
|
||||
"""Auto-seed default prompts if table is empty."""
|
||||
count = query("SELECT COUNT(*) as cnt FROM prompts")
|
||||
if count[0]["cnt"] > 0:
|
||||
return
|
||||
|
||||
execute("""
|
||||
INSERT INTO prompts (role, name, body, is_active, notes) VALUES (
|
||||
'extract', 'default-v1',
|
||||
'Ты — анализатор договоров облачного провайдера.\n\nНиже текст спецификации услуг из ПЕРВОГО документа (базовый договор).\nИзвлеки ВСЕ строки спецификации в JSON-массив.\n\nВерни СТРОГО JSON без пояснений. Не используй markdown-блоки.\n\nФОРМАТ:\n{\n "mode": "partial",\n "ops": [\n {"action": "ADD", "new_row": {"name": "полное наименование", "price": число, "qty": число, "sum": число, "date_start": "YYYY-MM-DD"}, "comment": ""}\n ]\n}\n\nПРАВИЛА:\n1. Извлеки КАЖДУЮ строку таблицы спецификации как отдельную ADD-операцию.\n2. Пропускай итоговые строки ("Итого...") и строки с подписями.\n3. Если ячейка пустая — ставь null (не пиши 0).\n4. price, qty, sum — ЧИСЛА, не строки.\n\nТЕКСТ ДОКУМЕНТА:\n---\n{doc_text}\n---',
|
||||
true, 'Авто-создан из llm_prompt.py _build_initial'
|
||||
)
|
||||
""")
|
||||
execute("""
|
||||
INSERT INTO prompts (role, name, body, is_active, notes) VALUES (
|
||||
'diff', 'default-v1',
|
||||
'Ты — анализатор допсоглашений к договорам облачного провайдера.\n\nУ тебя есть текущая спецификация услуг и текст нового допсоглашения (ДС).\nТвоя задача — определить, какие изменения вносит ДС в текущую спецификацию.\n\nВерни СТРОГО JSON без пояснений. Не используй markdown-блоки.\n\nФОРМАТ:\n{\n "mode": "partial" | "full_replace",\n "ops": [\n {"action": "ADD", "new_row": {"name": "...", "price": число, "qty": число, "sum": число, "date_start": "YYYY-MM-DD"}, "comment": "..."},\n {"action": "UPDATE", "target_id": "rN", "new_values": {"price": число}, "comment": "..."},\n {"action": "DELETE", "target_id": "rN", "comment": "..."},\n {"action": "UNRESOLVED", "new_values": {"name": "...", "price": число, ...}, "reason": "почему не смог сопоставить"}\n ]\n}\n\nПРАВИЛА:\n1. mode = "full_replace" — если в тексте есть фразы: «в следующей редакции», «заменить приложение», «излагается в следующей редакции». При full_replace — опиши ВСЕ новые строки как ADD.\n2. mode = "partial" — если ДС меняет только отдельные строки.\n3. Для UPDATE/DELETE — укажи target_id (r1, r2...) из списка текущей спецификации. НЕ придумывай новые id.\n4. new_values в UPDATE — только ИЗМЕНЁННЫЕ поля (не все).\n5. Если не можешь однозначно сопоставить строку — action: UNRESOLVED с reason.\n6. Пропускай итоговые строки и подписи.\n7. price, qty, sum — ЧИСЛА (не строки).\n\nТЕКУЩАЯ СПЕЦИФИКАЦИЯ:\n{spec_current}\n\nТЕКСТ ДОПСОГЛАШЕНИЯ:\n---\n{doc_text}\n---',
|
||||
true, 'Авто-создан из llm_prompt.py _build_diff'
|
||||
)
|
||||
""")
|
||||
@@ -0,0 +1,19 @@
|
||||
"""spec_current — текущее состояние спецификации."""
|
||||
from .connection import query
|
||||
|
||||
|
||||
def list_by_contract(contract_id):
|
||||
"""Return list of dicts with name_hash, name, price, qty, sum, date_start."""
|
||||
return query(
|
||||
"""SELECT name_hash, name, price, qty, sum, date_start
|
||||
FROM spec_current WHERE contract_id = %s ORDER BY name""",
|
||||
(contract_id,),
|
||||
)
|
||||
|
||||
|
||||
def get_elements_json(document_id):
|
||||
"""Get elements_json for a document."""
|
||||
rows = query(
|
||||
"SELECT elements_json FROM documents WHERE id = %s", (document_id,)
|
||||
)
|
||||
return rows[0]["elements_json"] if rows and rows[0]["elements_json"] else None
|
||||
@@ -0,0 +1,145 @@
|
||||
"""spec_events — event sourcing: apply ops, reset contract."""
|
||||
import json, uuid
|
||||
from .connection import query, execute
|
||||
|
||||
|
||||
def reset(contract_id):
|
||||
"""Clear spec_events + spec_current for contract."""
|
||||
execute("DELETE FROM spec_current WHERE contract_id = %s", (contract_id,))
|
||||
execute("DELETE FROM spec_events WHERE contract_id = %s", (contract_id,))
|
||||
|
||||
|
||||
def get_next_seq(contract_id):
|
||||
rows = query(
|
||||
"SELECT COALESCE(MAX(seq), 0) + 1 AS n FROM spec_events WHERE contract_id = %s",
|
||||
(contract_id,),
|
||||
)
|
||||
return rows[0]["n"] if rows else 1
|
||||
|
||||
|
||||
def apply_ops(contract_id, supplement_id, document_id, ops, prompt_id, raw_llm_response):
|
||||
"""Apply ADD/UPDATE/DELETE ops. Returns summary dict."""
|
||||
added = 0
|
||||
updated = 0
|
||||
deleted = 0
|
||||
seq = get_next_seq(contract_id)
|
||||
|
||||
for op in ops:
|
||||
action = op.get("action", "").upper()
|
||||
|
||||
if action == "ADD":
|
||||
nr = op.get("new_row", {})
|
||||
name = nr.get("name", "")
|
||||
name_hash = _hash(name)
|
||||
execute(
|
||||
"""INSERT INTO spec_events (contract_id, supplement_id, seq, action, target_hash,
|
||||
new_values, comment, status, prompt_version, source_document_id, raw_llm_response)
|
||||
VALUES (%s, %s, %s, 'ADD', %s, %s, %s, 'applied', %s, %s, %s)""",
|
||||
(
|
||||
contract_id, supplement_id, seq, name_hash,
|
||||
json.dumps(nr, ensure_ascii=False),
|
||||
op.get("comment", ""), prompt_id, document_id,
|
||||
json.dumps(raw_llm_response, ensure_ascii=False),
|
||||
),
|
||||
)
|
||||
_upsert_spec_current(contract_id, name_hash, nr)
|
||||
seq += 1
|
||||
added += 1
|
||||
|
||||
elif action == "UPDATE":
|
||||
nv = op.get("new_values", {})
|
||||
th = op.get("target_hash", "")
|
||||
execute(
|
||||
"""INSERT INTO spec_events (contract_id, supplement_id, seq, action, target_hash,
|
||||
new_values, comment, status, prompt_version, source_document_id, raw_llm_response)
|
||||
VALUES (%s, %s, %s, 'UPDATE', %s, %s, %s, 'applied', %s, %s, %s)""",
|
||||
(
|
||||
contract_id, supplement_id, seq, th,
|
||||
json.dumps(nv, ensure_ascii=False),
|
||||
op.get("comment", ""), prompt_id, document_id,
|
||||
json.dumps(raw_llm_response, ensure_ascii=False),
|
||||
),
|
||||
)
|
||||
_update_spec_current(contract_id, th, nv)
|
||||
seq += 1
|
||||
updated += 1
|
||||
|
||||
elif action == "DELETE":
|
||||
th = op.get("target_hash", "")
|
||||
execute(
|
||||
"""INSERT INTO spec_events (contract_id, supplement_id, seq, action, target_hash,
|
||||
new_values, comment, status, prompt_version, source_document_id, raw_llm_response)
|
||||
VALUES (%s, %s, %s, 'DELETE', %s, %s, %s, 'applied', %s, %s, %s)""",
|
||||
(
|
||||
contract_id, supplement_id, seq, th,
|
||||
json.dumps({}), op.get("comment", ""),
|
||||
prompt_id, document_id,
|
||||
json.dumps(raw_llm_response, ensure_ascii=False),
|
||||
),
|
||||
)
|
||||
execute("DELETE FROM spec_current WHERE contract_id = %s AND name_hash = %s", (contract_id, th))
|
||||
seq += 1
|
||||
deleted += 1
|
||||
|
||||
elif action == "UNRESOLVED":
|
||||
# Log but don't apply
|
||||
execute(
|
||||
"""INSERT INTO spec_events (contract_id, supplement_id, seq, action, target_hash,
|
||||
new_values, comment, status, prompt_version, source_document_id, raw_llm_response)
|
||||
VALUES (%s, %s, %s, 'UNRESOLVED', %s, %s, %s, 'unresolved', %s, %s, %s)""",
|
||||
(
|
||||
contract_id, supplement_id, seq,
|
||||
op.get("target_hash", ""),
|
||||
json.dumps(op.get("new_values", {}), ensure_ascii=False),
|
||||
op.get("reason", op.get("comment", "")),
|
||||
prompt_id, document_id,
|
||||
json.dumps(raw_llm_response, ensure_ascii=False),
|
||||
),
|
||||
)
|
||||
seq += 1
|
||||
|
||||
return {"added": added, "updated": updated, "deleted": deleted}
|
||||
|
||||
|
||||
def _hash(name):
|
||||
import hashlib
|
||||
return hashlib.sha256(name.strip().lower().encode()).hexdigest()[:16]
|
||||
|
||||
|
||||
def _upsert_spec_current(contract_id, name_hash, row):
|
||||
"""INSERT or UPDATE spec_current."""
|
||||
existing = query(
|
||||
"SELECT id FROM spec_current WHERE contract_id = %s AND name_hash = %s",
|
||||
(contract_id, name_hash),
|
||||
)
|
||||
if existing:
|
||||
execute(
|
||||
"""UPDATE spec_current SET name=%s, price=%s, qty=%s, sum=%s, date_start=%s, updated_at=now()
|
||||
WHERE contract_id=%s AND name_hash=%s""",
|
||||
(row.get("name"), row.get("price"), row.get("qty"), row.get("sum"),
|
||||
row.get("date_start"), contract_id, name_hash),
|
||||
)
|
||||
else:
|
||||
execute(
|
||||
"""INSERT INTO spec_current (contract_id, name_hash, name, price, qty, sum, date_start)
|
||||
VALUES (%s, %s, %s, %s, %s, %s, %s)""",
|
||||
(contract_id, name_hash, row.get("name"), row.get("price"),
|
||||
row.get("qty"), row.get("sum"), row.get("date_start")),
|
||||
)
|
||||
|
||||
|
||||
def _update_spec_current(contract_id, name_hash, new_values):
|
||||
"""Update specific fields in spec_current."""
|
||||
sets = []
|
||||
params = []
|
||||
for field in ("name", "price", "qty", "sum", "date_start"):
|
||||
if field in new_values:
|
||||
sets.append(f"{field} = %s")
|
||||
params.append(new_values[field])
|
||||
if sets:
|
||||
sets.append("updated_at = now()")
|
||||
params.extend([contract_id, name_hash])
|
||||
execute(
|
||||
f"UPDATE spec_current SET {', '.join(sets)} WHERE contract_id = %s AND name_hash = %s",
|
||||
params,
|
||||
)
|
||||
@@ -0,0 +1,47 @@
|
||||
"""Supplements CRUD."""
|
||||
from .connection import query, execute, execute_returning
|
||||
|
||||
|
||||
def insert(contract_id, document_id, supp_type="additional"):
|
||||
return execute_returning(
|
||||
"""INSERT INTO supplements (contract_id, document_id, type)
|
||||
VALUES (%s, %s, %s) RETURNING *""",
|
||||
(contract_id, document_id, supp_type),
|
||||
)
|
||||
|
||||
|
||||
def list_by_contract(contract_id):
|
||||
"""Supplements with parsed documents, ordered by created_at."""
|
||||
return query(
|
||||
"""SELECT s.id, s.type, s.document_id, d.filename
|
||||
FROM supplements s
|
||||
JOIN documents d ON s.document_id = d.id
|
||||
WHERE s.contract_id = %s AND d.elements_json IS NOT NULL
|
||||
ORDER BY s.created_at""",
|
||||
(contract_id,),
|
||||
)
|
||||
|
||||
|
||||
def get(supp_id):
|
||||
rows = query("SELECT * FROM supplements WHERE id = %s", (supp_id,))
|
||||
return rows[0] if rows else None
|
||||
|
||||
|
||||
def delete_by_document(contract_id, filename):
|
||||
"""Delete supplement+document by contract+filename."""
|
||||
row = query(
|
||||
"""SELECT s.id as sid, s.document_id FROM supplements s
|
||||
JOIN documents d ON d.id = s.document_id
|
||||
WHERE s.contract_id = %s AND d.filename = %s""",
|
||||
(contract_id, filename),
|
||||
)
|
||||
if row:
|
||||
r = row[0]
|
||||
execute("DELETE FROM supplements WHERE id = %s", (r["sid"],))
|
||||
execute("DELETE FROM documents WHERE id = %s", (r["document_id"],))
|
||||
return True
|
||||
return False
|
||||
|
||||
|
||||
def delete(supp_id):
|
||||
return execute("DELETE FROM supplements WHERE id = %s", (supp_id,))
|
||||
@@ -0,0 +1,37 @@
|
||||
"""LLM service — call LLM API."""
|
||||
import os, json
|
||||
import httpx
|
||||
|
||||
LLM_URL = "https://api.aillm.ru/v1/chat/completions"
|
||||
LLM_KEY = os.environ.get("LLM_KEY", "")
|
||||
LLM_MODEL = "gpt-oss-120b"
|
||||
|
||||
|
||||
def call_llm(current_spec, doc_text, build_prompt_fn):
|
||||
"""Call LLM. Returns (parsed_result, prompt_id)."""
|
||||
prompt, prompt_id = build_prompt_fn(current_spec, doc_text)
|
||||
payload = {
|
||||
"model": LLM_MODEL,
|
||||
"messages": [{"role": "user", "content": prompt}],
|
||||
"max_tokens": 8000,
|
||||
"temperature": 0.1,
|
||||
}
|
||||
with httpx.Client(http2=True, timeout=120, verify=False) as client:
|
||||
resp = client.post(
|
||||
LLM_URL,
|
||||
json=payload,
|
||||
headers={
|
||||
"Authorization": f"Bearer {LLM_KEY}",
|
||||
"Content-Type": "application/json",
|
||||
},
|
||||
)
|
||||
resp.raise_for_status()
|
||||
data = resp.json()
|
||||
|
||||
raw_text = data.get("choices", [{}])[0].get("message", {}).get("content", "")
|
||||
json_text = raw_text
|
||||
if "```json" in json_text:
|
||||
json_text = json_text.split("```json")[1].split("```")[0]
|
||||
elif "```" in json_text:
|
||||
json_text = json_text.split("```")[1].split("```")[0]
|
||||
return json.loads(json_text.strip()), prompt_id
|
||||
@@ -0,0 +1,73 @@
|
||||
"""Parse service — PDF (PyPDF2), DOCX (python-docx), plain text fallback."""
|
||||
import io
|
||||
from PyPDF2 import PdfReader
|
||||
|
||||
|
||||
def parse_file(filename, data):
|
||||
"""Parse file bytes → {status, elements, element_count}."""
|
||||
ext = filename.rsplit(".", 1)[-1].lower() if "." in filename else ""
|
||||
try:
|
||||
if ext == "pdf":
|
||||
return _parse_pdf(data)
|
||||
elif ext == "docx":
|
||||
return _parse_docx(data)
|
||||
elif ext == "doc":
|
||||
return _parse_docx(data) # python-docx handles most .doc
|
||||
else:
|
||||
return _parse_text(data)
|
||||
except Exception as e:
|
||||
return {"status": "error", "error": str(e), "element_count": 0}
|
||||
|
||||
|
||||
def _parse_pdf(data):
|
||||
reader = PdfReader(io.BytesIO(data))
|
||||
elements = []
|
||||
for page in reader.pages:
|
||||
text = page.extract_text()
|
||||
if text:
|
||||
for line in text.split("\n"):
|
||||
line = line.strip()
|
||||
if line:
|
||||
elements.append({"type": "paragraph", "text": line, "style": ""})
|
||||
return {"status": "parsed", "element_count": len(elements), "elements": elements}
|
||||
|
||||
|
||||
def _parse_docx(data):
|
||||
import docx
|
||||
doc = docx.Document(io.BytesIO(data))
|
||||
elements = []
|
||||
|
||||
for block in doc.element.body:
|
||||
tag = block.tag.split("}")[-1] if "}" in block.tag else block.tag
|
||||
if tag == "p":
|
||||
text = _extract_paragraph_text(block)
|
||||
if text:
|
||||
elements.append({"type": "paragraph", "text": text, "style": ""})
|
||||
elif tag == "tbl":
|
||||
rows = []
|
||||
for tr in block.findall(".//{http://schemas.openxmlformats.org/wordprocessingml/2006/main}tr"):
|
||||
cells = []
|
||||
for tc in tr.findall(".//{http://schemas.openxmlformats.org/wordprocessingml/2006/main}tc"):
|
||||
cell_text = "".join(t.text or "" for t in tc.findall(".//{http://schemas.openxmlformats.org/wordprocessingml/2006/main}t"))
|
||||
cells.append(cell_text.strip())
|
||||
if cells:
|
||||
rows.append(cells)
|
||||
if rows:
|
||||
elements.append({"type": "table", "rows": rows})
|
||||
|
||||
return {"status": "parsed", "element_count": len(elements), "elements": elements}
|
||||
|
||||
|
||||
def _extract_paragraph_text(p_elem):
|
||||
texts = []
|
||||
for t in p_elem.findall(".//{http://schemas.openxmlformats.org/wordprocessingml/2006/main}t"):
|
||||
if t.text:
|
||||
texts.append(t.text)
|
||||
return "".join(texts).strip()
|
||||
|
||||
|
||||
def _parse_text(data):
|
||||
text = data.decode("utf-8", errors="ignore")[:5000]
|
||||
lines = [l.strip() for l in text.split("\n") if l.strip()]
|
||||
return {"status": "parsed", "element_count": len(lines),
|
||||
"elements": [{"type": "paragraph", "text": l, "style": ""} for l in lines]}
|
||||
@@ -0,0 +1,132 @@
|
||||
"""Process service — SSE pipeline: reset → supplements → LLM → apply."""
|
||||
import json, time, threading
|
||||
from db import supplements, spec_current, spec_events
|
||||
|
||||
|
||||
def run_pipeline(contract_id, order_ids, sse_send, build_prompt_fn):
|
||||
"""Run full SSE comparison pipeline. sse_send is a callback(dict)."""
|
||||
t0 = time.time()
|
||||
|
||||
# 0. Reset
|
||||
spec_events.reset(contract_id)
|
||||
|
||||
# 1. Get supplements with parsed documents
|
||||
supps = supplements.list_by_contract(contract_id)
|
||||
if order_ids:
|
||||
order_list = [x.strip() for x in order_ids.split(",") if x.strip()]
|
||||
order_map = {oid: i for i, oid in enumerate(order_list)}
|
||||
supps.sort(key=lambda s: order_map.get(s["id"], 999999))
|
||||
|
||||
if not supps:
|
||||
sse_send({"type": "error", "message": "Нет распарсенных файлов"})
|
||||
return
|
||||
|
||||
from .llm import call_llm
|
||||
|
||||
for s in supps:
|
||||
sid = s["id"]
|
||||
|
||||
# Current spec
|
||||
cur = spec_current.list_by_contract(contract_id)
|
||||
current_spec = []
|
||||
for r in cur:
|
||||
current_spec.append({
|
||||
"hash": r["name_hash"],
|
||||
"name": r["name"],
|
||||
"price": float(r["price"]) if r.get("price") is not None else None,
|
||||
"qty": float(r["qty"]) if r.get("qty") is not None else None,
|
||||
"sum": float(r["sum"]) if r.get("sum") is not None else None,
|
||||
"date_start": r["date_start"],
|
||||
})
|
||||
|
||||
# Get elements_json
|
||||
ej = spec_current.get_elements_json(s["document_id"])
|
||||
if not ej:
|
||||
continue
|
||||
|
||||
if isinstance(ej, dict) and "Value" in ej:
|
||||
ej = ej["Value"]
|
||||
if isinstance(ej, str):
|
||||
elements = json.loads(ej)
|
||||
elif isinstance(ej, list):
|
||||
elements = ej
|
||||
else:
|
||||
elements = []
|
||||
|
||||
elements = [{k.lower(): v for k, v in el.items()} for el in elements]
|
||||
doc_text = _elements_to_text(elements)
|
||||
|
||||
sse_send({"type": "extract_start", "supplement_id": sid, "filename": s["filename"]})
|
||||
|
||||
# LLM call in thread with keepalive
|
||||
t1 = time.time()
|
||||
done = threading.Event()
|
||||
result = [None]
|
||||
error = [None]
|
||||
|
||||
def do_llm():
|
||||
try:
|
||||
result[0] = call_llm(current_spec, doc_text, build_prompt_fn)
|
||||
except Exception as e:
|
||||
error[0] = str(e)
|
||||
finally:
|
||||
done.set()
|
||||
|
||||
t = threading.Thread(target=do_llm)
|
||||
t.start()
|
||||
|
||||
while not done.wait(15):
|
||||
sse_send({"type": "keepalive"})
|
||||
|
||||
t.join()
|
||||
elapsed = round(time.time() - t1, 1)
|
||||
|
||||
if error[0]:
|
||||
sse_send({"type": "extract_error", "supplement_id": sid,
|
||||
"filename": s["filename"], "error": error[0], "time_s": elapsed})
|
||||
continue
|
||||
|
||||
llm_result, prompt_id = result[0]
|
||||
ops = llm_result.get("ops", [])
|
||||
mode = llm_result.get("mode", "partial")
|
||||
|
||||
# Enrich ops
|
||||
id_map = {f"r{i+1}": r["hash"] for i, r in enumerate(current_spec)}
|
||||
spec_names = {r["hash"]: r["name"] for r in current_spec}
|
||||
for op in ops:
|
||||
tid = op.get("target_id")
|
||||
if tid and tid in id_map:
|
||||
op["target_hash"] = id_map[tid]
|
||||
th = op.get("target_hash")
|
||||
if th and th in spec_names and not op.get("new_row", {}).get("name"):
|
||||
op.setdefault("new_row", {})["name"] = spec_names[th]
|
||||
|
||||
sse_send({"type": "llm_done", "supplement_id": sid, "filename": s["filename"],
|
||||
"ops_count": len(ops), "mode": mode, "time_s": elapsed})
|
||||
|
||||
# Apply events
|
||||
summary = spec_events.apply_ops(
|
||||
contract_id, sid, s.get("document_id", ""), ops, prompt_id, llm_result
|
||||
)
|
||||
sse_send({"type": "applied", "supplement_id": sid, "summary": summary, "ops": ops})
|
||||
|
||||
total_time = round(time.time() - t0, 1)
|
||||
sse_send({"type": "done", "total_time_s": total_time})
|
||||
|
||||
|
||||
def _elements_to_text(elements):
|
||||
lines = []
|
||||
for el in elements:
|
||||
if el.get("type") == "paragraph":
|
||||
prefix = f"[{el['style']}] " if el.get("style") else ""
|
||||
lines.append(prefix + el.get("text", ""))
|
||||
elif el.get("type") == "table":
|
||||
rows = el.get("rows", [])
|
||||
if rows:
|
||||
ncols = len(rows[0])
|
||||
lines.append(f"--- Таблица ({len(rows)}×{ncols}) ---")
|
||||
for row in rows:
|
||||
cells = [str(c).replace("\n", " ").replace("|", "\\|") for c in row[:ncols]]
|
||||
lines.append("| " + " | ".join(cells) + " |")
|
||||
lines.append("")
|
||||
return "\n".join(lines)
|
||||
@@ -0,0 +1,38 @@
|
||||
"""Unzip service — extract files from ZIP archive."""
|
||||
import zipfile, io, base64, json
|
||||
|
||||
|
||||
def handle_unzip(rfile, content_length):
|
||||
"""Parse ZIP upload, return list of extracted files."""
|
||||
body = rfile.read(content_length)
|
||||
|
||||
# Find file data in raw multipart
|
||||
# Simple approach: find ZIP bytes after header
|
||||
idx = body.find(b"\r\n\r\n")
|
||||
if idx < 0:
|
||||
return {"ok": False, "error": "invalid multipart"}
|
||||
|
||||
# Find first boundary end
|
||||
raw = body[idx + 4:]
|
||||
# Find the actual ZIP data (after Content-Disposition etc)
|
||||
zip_start = raw.find(b"PK\x03\x04")
|
||||
if zip_start < 0:
|
||||
return {"ok": False, "error": "not a ZIP file"}
|
||||
|
||||
zip_data = raw[zip_start:]
|
||||
|
||||
files = []
|
||||
with zipfile.ZipFile(io.BytesIO(zip_data)) as zf:
|
||||
for name in zf.namelist():
|
||||
if name.endswith("/"):
|
||||
continue
|
||||
data = zf.read(name)
|
||||
ext = name.rsplit(".", 1)[-1].lower() if "." in name else ""
|
||||
files.append({
|
||||
"filename": name,
|
||||
"data_b64": base64.b64encode(data).decode(),
|
||||
"ext": ext,
|
||||
"size": len(data),
|
||||
})
|
||||
|
||||
return {"ok": True, "files": files}
|
||||
@@ -0,0 +1,79 @@
|
||||
"""Upload service — accept file, store to DB, parse."""
|
||||
import uuid, base64, json
|
||||
from email.parser import BytesParser
|
||||
from db import documents, contracts, supplements
|
||||
from .parse import parse_file
|
||||
|
||||
|
||||
def handle_upload(rfile, content_type, content_length):
|
||||
"""Parse multipart upload, store in DB, return result dict."""
|
||||
body = rfile.read(content_length)
|
||||
|
||||
msg = BytesParser().parsebytes(
|
||||
b"Content-Type: " + content_type.encode() + b"\r\n\r\n" + body
|
||||
)
|
||||
filename = None
|
||||
file_data = None
|
||||
contract_id = ""
|
||||
|
||||
if msg.is_multipart():
|
||||
for part in msg.get_payload():
|
||||
fn = part.get_filename()
|
||||
if fn:
|
||||
filename = fn
|
||||
file_data = part.get_payload(decode=True)
|
||||
else:
|
||||
disp = part.get("Content-Disposition", "")
|
||||
if 'name="contract_id"' in disp:
|
||||
contract_id = part.get_payload(decode=True).decode("utf-8", errors="ignore").strip()
|
||||
|
||||
if not filename or not file_data:
|
||||
return {"ok": False, "error": "no file in request"}
|
||||
|
||||
mime = _mime_for(filename)
|
||||
b64 = base64.b64encode(file_data).decode()
|
||||
|
||||
# Удалить старый файл с тем же именем в этом контракте
|
||||
if contract_id:
|
||||
supplements.delete_by_document(contract_id, filename)
|
||||
|
||||
doc = documents.insert(filename, mime, b64)
|
||||
|
||||
# Создать контракт если нет
|
||||
if not contract_id:
|
||||
from datetime import datetime
|
||||
now = datetime.now()
|
||||
c = contracts.insert(f"б/н {now.strftime('%Y%m%d')}-{now.strftime('%H%M')}")
|
||||
contract_id = c["id"]
|
||||
supp_type = "initial"
|
||||
else:
|
||||
supp_type = "additional"
|
||||
|
||||
supplements.insert(contract_id, doc["id"], supp_type)
|
||||
|
||||
# Парсинг на VM
|
||||
parsed = parse_file(filename, file_data)
|
||||
if parsed and parsed.get("status") == "parsed":
|
||||
documents.set_parsed(doc["id"], parsed["elements"])
|
||||
elif parsed and parsed.get("status") == "error":
|
||||
documents.set_error(doc["id"], parsed.get("error", "parse failed"))
|
||||
|
||||
return {
|
||||
"ok": True,
|
||||
"doc_id": doc["id"],
|
||||
"contract_id": contract_id,
|
||||
"filename": filename,
|
||||
"size": len(file_data),
|
||||
"parsed": parsed,
|
||||
}
|
||||
|
||||
|
||||
def _mime_for(filename):
|
||||
ext = filename.rsplit(".", 1)[-1].lower() if "." in filename else ""
|
||||
mime_map = {
|
||||
"pdf": "application/pdf",
|
||||
"docx": "application/vnd.openxmlformats-officedocument.wordprocessingml.document",
|
||||
"doc": "application/msword",
|
||||
"zip": "application/zip",
|
||||
}
|
||||
return mime_map.get(ext, "application/octet-stream")
|
||||
Reference in New Issue
Block a user