No answer being retrieved from "Retrieval testing"

[alspires]

Hello,

Another problem we've been having in the "Retrieval testing" tab. We are making questions that answers for it are contained in the uploaded document and that used to work on previous versions, but in this new version the testing returns empty. I have checked the logs, there are no errors and calls made to our service endpoints seems alright.

2025-11-03 10:19:39,367 INFO     46 127.0.0.1 - - [03/Nov/2025 10:19:39] "GET /v1/kb/d3778c5cb57e11f0826f0aa8c8eda76f/knowledge_graph HTTP/1.1" 200 -
2025-11-03 10:19:40,652 DEBUG    46 [TKS] creat data gateway sql login user
2025-11-03 10:19:40,653 DEBUG    46 [TKS]
2025-11-03 10:19:40,769 DEBUG    46 [TKS] datum
2025-11-03 10:19:40,769 DEBUG    46 [TKS]
2025-11-03 10:19:40,769 DEBUG    46 [TKS]
2025-11-03 10:19:40,770 DEBUG    46 [TKS]
2025-11-03 10:19:40,771 DEBUG    46 [TKS] drug user exploit
2025-11-03 10:19:41,222 DEBUG    46 Request options: {'method': 'post', 'url': '/embeddings', 'files': None, 'idempotency_key': 'stainless-python-retry-168840f5-0694-464d-b029-597871db4fc4', 'post_parser': <function Embeddings.create.<locals>.parser at 0x7fab80b1e9e0>, 'json_data': {'input': ['How to create a Data Gateways SQL login and user?'], 'model': 'all-mpnet-base-v2', 'encoding_format': 'float'}, 'extra_json': {'drop_params': True}}
2025-11-03 10:19:41,224 DEBUG    46 Sending HTTP Request: POST https://api-ai.cloud/v1/embeddings
2025-11-03 10:19:41,225 DEBUG    46 connect_tcp.started host='api-ai.cloud' port=443 local_address=None timeout=5.0 socket_options=None
2025-11-03 10:19:41,237 DEBUG    46 connect_tcp.complete return_value=<httpcore._backends.sync.SyncStream object at 0x7fab7d656aa0>
2025-11-03 10:19:41,237 DEBUG    46 start_tls.started ssl_context=<ssl.SSLContext object at 0x7fab8049cf40> server_hostname='api-ai.cloud' timeout=5.0
2025-11-03 10:19:41,250 DEBUG    46 start_tls.complete return_value=<httpcore._backends.sync.SyncStream object at 0x7fab7d656b60>
2025-11-03 10:19:41,250 DEBUG    46 send_request_headers.started request=<Request [b'POST']>
2025-11-03 10:19:41,251 DEBUG    46 send_request_headers.complete
2025-11-03 10:19:41,251 DEBUG    46 send_request_body.started request=<Request [b'POST']>
2025-11-03 10:19:41,251 DEBUG    46 send_request_body.complete
2025-11-03 10:19:41,251 DEBUG    46 receive_response_headers.started request=<Request [b'POST']>
2025-11-03 10:19:41,558 DEBUG    46 receive_response_headers.complete return_value=(b'HTTP/1.1', 200, b'OK', [(b'Date', b'Mon, 03 Nov 2025 10:19:41 GMT'), (b'Content-Type', b'application/json'), (b'Transfer-Encoding', b'chunked'), (b'Connection', b'keep-alive'), (b'Content-Encoding', b'gzip'), (b'Strict-Transport-Security', b'max-age=15552000; includeSubDomains'), (b'cf-cache-status', b'DYNAMIC'), (b'Set-Cookie', b'__cf_bm=EH_3SyJ2vUDnlUVgt.pILW_tGbSh3SdJvNNDj74PuxI-1762165181-1.0.1.1-NHTloXWKCh9DU4OkpAqcI4pIND5XLzW2Q8wPqlGF5ekUB1.dGs7ScPCvD0q2E57XZMq9RaYDZfVznCwNre.70qIlj69YX_9r0XxeDnuoDGY; path=/; expires=Mon, 03-Nov-25 10:49:41 GMT; domain=cloud; HttpOnly; Secure; SameSite=None'), (b'Server', b'cloudflare'), (b'CF-RAY', b'998b227edbb8a9d6-IAD')])
2025-11-03 10:19:41,559 DEBUG    46 receive_response_body.started request=<Request [b'POST']>
2025-11-03 10:19:41,561 DEBUG    46 receive_response_body.complete
2025-11-03 10:19:41,561 DEBUG    46 response_closed.started
2025-11-03 10:19:41,562 DEBUG    46 response_closed.complete

Could this be a configuration error, or could it be a bug?

Thanks.

[dosubot[bot]]

This looks like a bug introduced in recent versions. The retrieval pipeline now filters for chunks with available_int=1, but newly created document chunks are missing this field, so they’re silently excluded from results—even though everything else appears to work and logs show no errors. This is a known issue and has been reported by other users after upgrades as well [source].

To fix this, re-parse your documents or manually set the available_int flag to 1 for all chunks in your knowledge base. Most users resolve this by upgrading to the latest nightly (not nightly-slim) Docker image and re-parsing their documents [source]. Also, double-check that your knowledge base and embedding model settings match, and that your similarity threshold isn’t set too high.

If you still see empty results after these steps, check the backend and frontend logs for any warnings when accessing the retrieval testing page or uploading files. Let me know if you need details on how to check or update chunk flags.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}    

[Magicbook1108]

Duplicate with #11006