dify/api/tasks/document_indexing_update_task.py

import logging
import time

import click
from celery import shared_task
from sqlalchemy import delete, select

from core.db.session_factory import session_factory
from core.indexing_runner import DocumentIsPausedError, IndexingRunner
from core.rag.index_processor.constant.index_type import IndexStructureType, IndexTechniqueType
from core.rag.index_processor.index_processor_factory import IndexProcessorFactory
from libs.datetime_utils import naive_utc_now
from models.dataset import Dataset, Document, DocumentSegment
from models.enums import IndexingStatus
from tasks.generate_summary_index_task import generate_summary_index_task

logger = logging.getLogger(__name__)


@shared_task(queue="dataset")
def document_indexing_update_task(dataset_id: str, document_id: str):
    """
    Async update document
    :param dataset_id:
    :param document_id:

    Usage: document_indexing_update_task.delay(dataset_id, document_id)
    """
    logger.info(click.style(f"Start update document: {document_id}", fg="green"))
    start_at = time.perf_counter()

    with session_factory.create_session() as session, session.begin():
        document = session.scalar(
            select(Document).where(Document.id == document_id, Document.dataset_id == dataset_id).limit(1)
        )

        if not document:
            logger.info(click.style(f"Document not found: {document_id}", fg="red"))
            return

        document.indexing_status = IndexingStatus.PARSING
        document.processing_started_at = naive_utc_now()

        dataset = session.scalar(select(Dataset).where(Dataset.id == dataset_id).limit(1))
        if not dataset:
            return

        index_type = document.doc_form
        segments = session.scalars(select(DocumentSegment).where(DocumentSegment.document_id == document_id)).all()
        index_node_ids = [segment.index_node_id for segment in segments if segment.index_node_id]

    clean_success = False
    try:
        index_processor = IndexProcessorFactory(index_type).init_index_processor()
        if index_node_ids:
            index_processor.clean(dataset, index_node_ids, with_keywords=True, delete_child_chunks=True)
            end_at = time.perf_counter()
            logger.info(
                click.style(
                    "Cleaned document when document update data source or process rule: {} latency: {}".format(
                        document_id, end_at - start_at
                    ),
                    fg="green",
                )
            )
            clean_success = True
    except Exception:
        logger.exception("Failed to clean document index during update, document_id: %s", document_id)

    if clean_success:
        with session_factory.create_session() as session, session.begin():
            segment_delete_stmt = delete(DocumentSegment).where(DocumentSegment.document_id == document_id)
            session.execute(segment_delete_stmt)

    has_error = False
    try:
        indexing_runner = IndexingRunner()
        indexing_runner.run([document])
        end_at = time.perf_counter()
        logger.info(click.style(f"update document: {document.id} latency: {end_at - start_at}", fg="green"))
    except DocumentIsPausedError as ex:
        logger.info(click.style(str(ex), fg="yellow"))
        has_error = True
    except Exception:
        logger.exception("document_indexing_update_task failed, document_id: %s", document_id)
        has_error = True

    if has_error:
        return

    # Trigger summary index generation for the updated document if enabled.
    # Only generate for high_quality indexing technique and when summary_index_setting is enabled.
    with session_factory.create_session() as session:
        dataset = session.scalar(select(Dataset).where(Dataset.id == dataset_id).limit(1))
        if not dataset:
            logger.warning("Dataset %s not found after update indexing", dataset_id)
            return

        if dataset.indexing_technique == IndexTechniqueType.HIGH_QUALITY:
            summary_index_setting = dataset.summary_index_setting
            if summary_index_setting and summary_index_setting.get("enable"):
                session.expire_all()
                document = session.scalar(
                    select(Document).where(Document.id == document_id, Document.dataset_id == dataset_id).limit(1)
                )
                if (
                    document
                    and document.indexing_status == IndexingStatus.COMPLETED
                    and document.doc_form != IndexStructureType.QA_INDEX
                    and document.need_summary is True
                ):
                    try:
                        generate_summary_index_task.delay(dataset.id, document.id, None)
                        logger.info(
                            "Queued summary index generation task for document %s in dataset %s "
                            "after update indexing completed",
                            document.id,
                            dataset.id,
                        )
                    except Exception:
                        logger.exception(
                            "Failed to queue summary index generation task for document %s after update",
                            document.id,
                        )