dify/api/tasks/add_document_to_index_task.py

85 lines
2.7 KiB
Python
Raw Normal View History

2023-05-15 08:51:32 +08:00
import datetime
import logging
import time
import click
from celery import shared_task
from langchain.schema import Document
2023-05-15 08:51:32 +08:00
from werkzeug.exceptions import NotFound
from core.index.index import IndexBuilder
2023-05-15 08:51:32 +08:00
from extensions.ext_database import db
from extensions.ext_redis import redis_client
from models.dataset import DocumentSegment
from models.dataset import Document as DatasetDocument
2023-05-15 08:51:32 +08:00
@shared_task
def add_document_to_index_task(dataset_document_id: str):
2023-05-15 08:51:32 +08:00
"""
Async Add document to index
:param document_id:
Usage: add_document_to_index.delay(document_id)
"""
logging.info(click.style('Start add document to index: {}'.format(dataset_document_id), fg='green'))
2023-05-15 08:51:32 +08:00
start_at = time.perf_counter()
dataset_document = db.session.query(DatasetDocument).filter(DatasetDocument.id == dataset_document_id).first()
if not dataset_document:
2023-05-15 08:51:32 +08:00
raise NotFound('Document not found')
if dataset_document.indexing_status != 'completed':
2023-05-15 08:51:32 +08:00
return
indexing_cache_key = 'document_{}_indexing'.format(dataset_document.id)
2023-05-15 08:51:32 +08:00
try:
segments = db.session.query(DocumentSegment).filter(
DocumentSegment.document_id == dataset_document.id,
2023-05-15 08:51:32 +08:00
DocumentSegment.enabled == True
) \
.order_by(DocumentSegment.position.asc()).all()
documents = []
2023-05-15 08:51:32 +08:00
for segment in segments:
document = Document(
page_content=segment.content,
metadata={
"doc_id": segment.index_node_id,
"doc_hash": segment.index_node_hash,
"document_id": segment.document_id,
"dataset_id": segment.dataset_id,
}
2023-05-15 08:51:32 +08:00
)
documents.append(document)
2023-05-15 08:51:32 +08:00
dataset = dataset_document.dataset
2023-05-15 08:51:32 +08:00
if not dataset:
raise Exception('Document has no dataset')
# save vector index
index = IndexBuilder.get_index(dataset, 'high_quality')
if index:
index.add_texts(documents)
2023-05-15 08:51:32 +08:00
# save keyword index
index = IndexBuilder.get_index(dataset, 'economy')
if index:
index.add_texts(documents)
2023-05-15 08:51:32 +08:00
end_at = time.perf_counter()
logging.info(
click.style('Document added to index: {} latency: {}'.format(dataset_document.id, end_at - start_at), fg='green'))
2023-05-15 08:51:32 +08:00
except Exception as e:
logging.exception("add document to index failed")
dataset_document.enabled = False
dataset_document.disabled_at = datetime.datetime.utcnow()
dataset_document.status = 'error'
dataset_document.error = str(e)
2023-05-15 08:51:32 +08:00
db.session.commit()
finally:
redis_client.delete(indexing_cache_key)