[Bug?] "terminated" error with Ollama Embeddings + Qdrant toolVectorStore — no crash, no SIGTERM in logs

Questions
1

My workflow returns a "terminated" error when using the Embeddings Ollama node connected to a Qdrant Vector Store via toolVectorStore. The AI Agent node never calls the LLM connected to the tool.

The strange part: n8n does not crash, there is no SIGTERM in the container logs, and Qdrant correctly receives and responds to the query (HTTP 200).

What is the error message

{ "errorMessage": "terminated", "errorDescription": "terminated", "errorDetails": {}, "n8nDetails": { "n8nVersion": "2.18.4 (Self Hosted)", "binaryDataMode": "filesystem" } }

Workflow structure

What I’ve verified

  • Qdrant receives the query from n8n and responds correctly (POST /collections/.../points/query HTTP/1.1" 200)

  • docker inspect n8n --format='{{.State.OOMKilled}}' → false

  • docker logs n8n | grep -i "SIGTERM\|runner\|crash\|terminated" → no output

  • The workflow worked correctly until a few days ago — the only change was a manual docker pull n8nio/n8n:latest (upgrade to 2.18.4)

Information on your n8n setup

  • n8n version: 2.18.4 (Self Hosted)

  • Database: SQLite (default)

  • Running n8n via: Docker Compose

  • Operating system: Ubuntu (VPS)

  • Binary data mode: filesystem

  • Runners mode: internal (default)

  • NODE_OPTIONS: --max-old-space-size=4096

  • Memory limit: 5600M

Why does the execution show "terminated" with no trace in container logs?

2

@the-code-learner In n8n 2.18.4, "terminated" usually means the AI Agent killed the execution due to a timeout or toolVectorStore/Embeddings hang (not a crash or OOM).

3

How can I check it? It seems that n8n does not pass the query and the output of the rag to the AI that in this case is Gemma4E4B. I don’t see any log in n8n, llama.cpp or ollama. My main issue is that I can’t find enough details about this error to debug it

4

Check it like this:

  1. Enable full execution logs
N8N_LOG_LEVEL=debug
  1. Check execution data in UI

  • Open the failed run

  • Look for the last node that actually produced output (usually Embeddings or Qdrant)

  1. Inspect agent tool call timing

  • In execution view → “AI Agent” → check if tool call ever appears

  • If not → agent never reached tool stage

  1. Add manual checkpoints

  • Put a Set/Function node after:

    • Embeddings

    • Qdrant

  • See where execution stops

  1. Check timeout silently killing run
EXECUTIONS_TIMEOUT
EXECUTIONS_TIMEOUT_MAX

If Qdrant + Ollama have no logs, the stop is happening inside n8n AI Agent (not external services).

5

thank you for showing me how to enable full logs! I found the issue: I had an expression for the description of the tool, but its result was too long (3300 characters) and it seems it was breaking the tool itself. Now it is 600 characters and it works

6

I found out also another potential issue related to the qdrant search filter option. It was taking items from outside the loop potentially generating the error