Skip to content

armgit / dify

Public Imported from GitHub
0

Build Agentic workflows, RAG pipelines, with rich AI model and tool support on one collaborative workspace. Deploy on cloud, VPC, or self-hosted, so teams move from prototype to production without rebuilding the stack.

3.2 KB
Raw Blame
"""Persist summary indexing outcomes using one owned session per attempt."""

from collections.abc import Callable

from sqlalchemy import select
from sqlalchemy.orm import Session

from libs.datetime_utils import naive_utc_now
from models.dataset import Dataset, Document, DocumentSegment, DocumentSegmentSummary
from models.enums import SummaryStatus
from services.knowledge.summaries.application import SummaryVectorInput, SummaryVectorResult


class SQLAlchemySummaryRepository:
    def __init__(self, *, new_session: Callable[[], Session]) -> None:
        self._new_session = new_session

    @staticmethod
    def _record(session: Session, command: SummaryVectorInput) -> DocumentSegmentSummary | None:
        ref = command.ref
        segment = session.scalar(
            select(DocumentSegment)
            .join(Document, Document.id == DocumentSegment.document_id)
            .join(Dataset, Dataset.id == Document.dataset_id)
            .where(
                Dataset.id == ref.document.dataset.dataset_id,
                Dataset.tenant_id == ref.document.dataset.tenant_id,
                Document.id == ref.document.document_id,
                Document.tenant_id == ref.document.dataset.tenant_id,
                DocumentSegment.id == ref.segment_id,
                DocumentSegment.dataset_id == ref.document.dataset.dataset_id,
                DocumentSegment.tenant_id == ref.document.dataset.tenant_id,
            )
        )
        if segment is None:
            raise LookupError("Summary segment no longer exists in this dataset")
        return session.scalar(
            select(DocumentSegmentSummary)
            .where(
                DocumentSegmentSummary.dataset_id == ref.document.dataset.dataset_id,
                DocumentSegmentSummary.document_id == ref.document.document_id,
                DocumentSegmentSummary.chunk_id == ref.segment_id,
            )
            .limit(1)
        )

    def completed(self, command: SummaryVectorInput, result: SummaryVectorResult) -> None:
        with self._new_session() as session, session.begin():
            row = self._record(session, command)
            if row is None:
                row = DocumentSegmentSummary(
                    dataset_id=command.ref.document.dataset.dataset_id,
                    document_id=command.ref.document.document_id,
                    chunk_id=command.ref.segment_id,
                    summary_content=command.content,
                    enabled=True,
                )
                row.id = command.summary_id
                session.add(row)
            row.summary_content = command.content
            row.summary_index_node_id = result.node_id
            row.summary_index_node_hash = result.content_hash
            row.tokens = result.tokens
            row.status = SummaryStatus.COMPLETED
            row.error = None
            row.updated_at = naive_utc_now()

    def failed(self, command: SummaryVectorInput, error: str) -> None:
        with self._new_session() as session, session.begin():
            row = self._record(session, command)
            if row is not None:
                row.status = SummaryStatus.ERROR
                row.error = error
                row.updated_at = naive_utc_now()