How to Fix Knowledge Base Vector Search After Upgrading Dify

Discussion
1

How to Fix Knowledge Base Vector Search After Upgrading Dify | Dify 升级后知识库向量检索报错的修复方法

Problem | 问题

After upgrading Dify (e.g., from 1.8.x/1.10.x to 1.11.0+), you may encounter this error when testing knowledge base retrieval:

升级 Dify（例如从 1.8.x/1.10.x 升级到 1.11.0+）后，测试知识库检索时可能遇到以下错误：

Query call with protocol GRPC search failed with message extract target vectors: class 
Vector_index_xxx_Node does not have named vector default configured. 
Available named vectors map[].

Root Cause | 根本原因

English: This is caused by a Weaviate schema format change. Old knowledge bases use vectorIndexConfig at the top level, but newer Dify versions expect vectorConfig.default. This mismatch breaks vector search.

中文： 这是由 Weaviate schema 格式变更引起的。旧知识库在顶层使用 vectorIndexConfig，但新版 Dify 期望 vectorConfig.default 格式。这种不匹配导致向量搜索失败。

Solution | 解决方案

I’ve created a Python script to batch fix this issue:

我创建了一个 Python 脚本来批量修复此问题：

:link: GitHub Repository | 仓库地址

Quick Start | 快速开始

Step 1: Download and copy script | 下载并复制脚本

# Download | 下载
curl -o /tmp/batch_fix_weaviate.py https://raw.githubusercontent.com/yupoet/dify-weaviate-fix/main/batch_fix_weaviate.py

# Copy to container | 复制到容器
docker cp /tmp/batch_fix_weaviate.py docker-api-1:/tmp/

Step 2: Scan affected collections | 扫描受影响的 collections

docker exec docker-api-1 python /tmp/batch_fix_weaviate.py scan

Step 3: Fix all affected collections | 修复所有受影响的 collections

docker exec -it docker-api-1 python /tmp/batch_fix_weaviate.py fix

Step 4: Re-embed in Dify | 在 Dify 中重新嵌入

English: For each affected knowledge base, go to Settings → Embedding Model, switch to a different model, save, then switch back and save again. This forces re-embedding.

中文： 对每个受影响的知识库，进入 设置 → Embedding 模型，切换到另一个模型并保存，然后切换回原模型再保存。这会强制重新嵌入。

Available Commands | 可用命令

Command Description 说明
scan List collections needing fix 列出需要修复的 collections
dry-run Simulate fix (no changes) 模拟修复（不实际执行）
fix Fix all affected collections 修复所有受影响的 collections
fix-one <name> Fix single collection 修复单个 collection
cleanup Remove orphaned collections 删除孤立的 collections
list-names Show dataset names 显示知识库名称

Environment Variables | 环境变量

Variable Default Description 说明
WEAVIATE_ENDPOINT http://weaviate:8080 Weaviate URL Weaviate 地址
WEAVIATE_API_KEY (empty) API key API 密钥

Tested With | 测试环境

  • Dify 1.11.0

  • Weaviate 1.27.0

Full Documentation | 完整文档

If this helped you, please :star: the repository!

如果对你有帮助，请给仓库点个 :star:！

GitHub: GitHub - yupoet/dify-weaviate-fix: How to Fix Knowledge Base Vector Search After Upgrading Dify (Weaviate vectorConfig Issue)

2

There is already an official migration guide and a migration script available for this issue.

Additionally, a simplified community-based Gist of the official migration guide has also been made available.

3

Thanks for sharing the official resources! I’ve reviewed them carefully.

Clarification on the difference:

Official Solution My Solution
Approach Migrate data (preserve vectors) Rebuild schema + re-embed
Preserves vectors :white_check_mark: Yes :cross_mark: No, requires re-embedding
Best for Large datasets Small datasets or when switching embedding models anyway
Complexity Higher (LSM fix + data migration) Lower (just schema recreation)

My solution is intended as a quick fix for:

  • Small deployments with few knowledge bases
  • Users who want to switch embedding models anyway
  • Situations where re-embedding is acceptable

The official solution is better for:

  • Large deployments with many documents
  • Users who want to preserve existing vectors
  • Production environments where re-embedding would take too long

I’ll update my GitHub repo to clarify this positioning and add references to the official guide. Thanks for the feedback! :folded_hands:

感谢分享官方资源！我仔细查看了这些文档。

关于两种方案的区别说明：

我的方案定位是一个快速简化修复，适用于：

  • 知识库数量较少的小型部署
  • 本来就想切换 embedding 模型的用户
  • 可以接受重新嵌入的场景

官方方案更适合：

  • 文档量大的大型部署
  • 需要保留现有向量数据的用户
  • 重新嵌入耗时过长的生产环境

我会更新 GitHub 仓库，明确方案定位并添加官方指南的引用。感谢反馈！:folded_hands: