Describe the problem/error/question
The Qdrant Vector Store node suddenly stops returning data. It was working successfully and then after a handful of executions stops returning any results. It doesn’t fail, it just says no results returned.
The logs in Qdrant do not indicate any errors. I’m self-hosting Qdrant but memory and CPU are sufficient for the size of my collection. I can see I don’t break 30% usage on memory and <1% on CPU. I’ve restarted Qdrant but still the same issue.
I have used the embeddings generated from n8n to send a request to Qdrant’s API and I get results back so I don’t think Qdrant is the issue.
curl --request POST \
--url http://192.168.1.35:6333/collections/brennan-work-kb/points/search \
--header 'content-type: application/json' \
--data '{
"vector": [0.02645673044025898, 0.004150944296270609, ...],
"with_vectors": false,
"with_payload": true,
"limit": 1
}'
I’m not even sure how to troubleshoot this since I don’t know where I can find any additional detail to investigate.
EDIT It’s the same situation if I use Qdrant Cloud so the issue is not specific to my self-hosted version of Qdrant.
What is the error message (if any)?
No error message. The node says it executed successfully but no data returned. No errors in the console. No errors in the Qdrant logs.
Please share your workflow
Share the output returned by the last node
It’s empty - it doesn’t return any results but it should.
Information on your n8n setup
- n8n version: 1.91.2
- Database (default: SQLite): SQLite
- n8n EXECUTIONS_PROCESS setting (default: own, main): main?
- Running n8n via (Docker, npm, n8n cloud, desktop app): Docker
- Operating system: Ubuntu
Tried inserting exactly the query and then pulled it back with no problem. If this query is not returning anything, there must be something wrong with the points in the DB.
I was getting results from n8n for a handful of calls but then n8n stops returning anything. I can get results via a REST API call to Qdrant so I’m not sure how to evaluate problems with the points in the DB.
The points in the DB are loaded via a different n8n workflow but they use the same embedding model and number of dimensions.
Do you have any suggestions for how I might troubleshoot that?
What if you remove any prompt and run it, do you get any results at all?
in the executions, do you see that the embedding model generated the vector?
Yes, I can see the embeddings were generated. I’ve used those in the curl request to Qdrant’s API and successfully returned responses so I know the embeddings are fine.
n8n has been sitting untouched with no workflows running for about two hours. I tried it without the prompt and got results. I added the prompt back and I’m getting results again.
I turned on “Include Metadata” and stopped getting results. When I toggled it back off, I’m still not getting results.
It’s the same thing with both Qdrant Cloud and the self-hosted version so not isolated to my self-hosted version of the database.
I’ve gotten results with “Include Metadata” toggled before so I don’t know why that would create an issue suddenly. Is there anywhere I can get a better view into n8n logs?
Another wrinkle - If I “Test Execute” my workflow (red box) while viewing the entire workflow, it works.
I can open the node to see results returned. It works even when I include the query or even when I toggle “Include Metadata”.
However, if I go into the Qdrant node in the workflow and try to test only that node, it returns no results. I did not change anything from the successful test run.
Sorry, I wish I could repro this in my environment, but I can’t, every request, sent from either the node execution or workflow execution just works.
If you think this is a bug, please file one here.
No worries. Appreciate the assistance… it at least helped me see some different behavior that might help as I investigate further.
