SIGN IN SIGN UP
infiniflow / ragflow UNCLAIMED

RAGFlow is a leading open-source Retrieval-Augmented Generation (RAG) engine that fuses cutting-edge RAG with Agent capabilities to create a superior context layer for LLMs

0 0 176 Python

Validate chunk image_base64 before doc-store write (#15364)

## Summary

Fixes [#15363](https://github.com/infiniflow/ragflow/issues/15363) —
`add_chunk` / `update_chunk` indexed chunks with `image_id` before
validating or storing `image_base64`, leaving orphan chunks on invalid
input.

## Related Issue

Fixes #15363

## Change Type

- [x] Bug fix
- [x] Regression tests

## What Changed

- Added `_decode_chunk_image_base64()` — strict base64 decode with
structured 4xx errors
- Added `_store_chunk_image_or_error()` — catches `store_chunk_image`
failures
- **`add_chunk` / `update_chunk`**: decode + store image **before**
`docStoreConn.insert` / `update`; only set `img_id` after successful
storage

## Files Changed

| File | Change |
|------|--------|
| `api/apps/restful_apis/chunk_api.py` | Helpers + reorder image
handling |
| `test/testcases/test_web_api/test_chunk_app/test_chunk_routes_unit.py`
| 3 regression tests |

## Validation

```bash
cd /root/gittensor/ragflow
pytest test/testcases/test_web_api/test_chunk_app/test_chunk_routes_unit.py::test_restful_add_chunk_invalid_image_base64_does_not_index_chunk -v
pytest test/testcases/test_web_api/test_chunk_app/test_chunk_routes_unit.py::test_restful_update_chunk_invalid_image_base64_does_not_update_chunk -v
pytest test/testcases/test_web_api/test_chunk_app/test_chunk_routes_unit.py::test_restful_add_chunk_valid_image_base64_stores_before_insert -v
pytest test/testcases/test_web_api/test_chunk_app/test_chunk_routes_unit.py -v
```

## Test Plan

- [x] Invalid `image_base64` on add → 4xx, no doc-store insert
- [x] Invalid `image_base64` on update → 4xx, no doc-store update
- [x] Valid PNG base64 on add → image stored, chunk indexed with
`img_id`
- [ ] CI green
K
kpdev committed
cb1ea5a47fe2b93f5c8e432be98ae40c34c47cfb
Parent: 04aa8d0
Committed by GitHub <noreply@github.com> on 5/29/2026, 11:36:46 AM