GitHub Repo
Open WebUI Community Reports a Qdrant Write Race That Can Leave Knowledge Bases Missing Document Chunks
An Open WebUI community report dated October 6 says the knowledge-base workflow may read from Qdrant before a batch write finishes, leaving an incomplete copy of a document. The report includes version details, logs, and a verification method, but no maintainer confirmation or merged fix has been reported.
An integrity issue in Open WebUI’s document import workflow merits attention from RAG engineers. An issue filed on October 6 says that when using the Qdrant backend, a document may show as completed and the API may return success even though the knowledge base receives only some of its chunks. The reporter observed the problem in v0.11.4 and confirmed by reading the code that the development branch retained the same code path; they did not reproduce it on a development build. Community report
The issue involves a two-stage data transfer: the system first writes chunks to a document collection, then reads them back and adds them to the knowledge base. The reporter says the upload_points call does not specify a wait parameter, so it uses the default of 64 points per batch and wait=False. The next step may begin reading before all batches have been applied. In the reported logs, the document had 86 chunks, but the read returned only 64. Workflow and logs
Qdrant’s official documentation confirms the API semantics: a response with wait=false means the request was received, while the data may still be processing in the background; it does not guarantee that the data is immediately readable. Applications that need to proceed only after the operation finishes can use wait=true. This supports the proposed race mechanism, but does not independently verify the Open WebUI bug. Qdrant documentation
For RAG, this kind of omission could make some evidence unavailable to retrieval. That risk is inferred from the data flow; no answer-quality evaluation has been reported. Engineers should verify document processing status separately from knowledge-base content integrity, especially when synchronizing immediately after a document update or importing documents in batches.
The reporter recommends waiting for writes to finish and comparing the chunk contents in the source and knowledge-base copies. However, re-chunking can change the number of chunks, so a count difference alone cannot establish that data was lost. As of the time of review, the issue remained open and listed no fix PR. Follow-up should confirm whether a fix covers both Qdrant clients and assess the effect of waiting for completion on import latency. Recommendations and status