🆘 Critical: Single Frontend Request triggers Multiple Parallel Executions (Ghost Retries) on ~10s Workflow

Help me Build my Workflow
1

Hi everyone,

I am finalizing a Concierge App for a client and facing a critical stability issue I can’t seem to shake. The workflow logic is perfect, but the execution behavior is unstable.

:pushpin: The Problem When I trigger the workflow once from my frontend (React), n8n starts 5 to 7 parallel executions for that single request. Even though I optimized the workflow down to ~10 seconds latency, these “ghost retries” keep happening, unnecessarily burning my API credits (OpenAI, etc.).

:gear: The Stack

  • Backend: n8n (Self-hosted).

  • Trigger: Webhook Node (Method: POST, Response Mode: Using 'Respond to Webhook' Node).

  • Frontend: React App (Bolt).

  • Workflow Logic: Google Sheets Lookup + LLM Analysis + JSON Response.

  • Current Execution Time: ~10 seconds (down from 20s after optimization).

:man_detective: Troubleshooting Steps Already Taken

I initially thought it was a browser timeout issue, so I hardened both sides:

1. Frontend (React)

  • UI Blocking: The submit button is disabled immediately upon clicking (preventing double-clicks).

  • Timeout Config: I explicitly set the fetch/axios timeout to 60,000ms (60s) to prevent the browser from cutting the connection prematurely.

  • Network Tab: The browser shows “Pending” for the first request, but n8n logs show new executions starting at t+2s, t+5s, etc.

2. Backend (n8n)

  • Optimization: I fixed previous Google Sheets “Quota Exceeded” errors by using “Execute Once”. The workflow is now green and efficient.

  • Workflow Settings: I checked for loops, but the execution logs show distinct new IDs starting from the Webhook node.

:red_question_mark: My Hypothesis & Question

Since the workflow takes ~10s to reach the “Respond to Webhook” node, I suspect an infrastructure timeout layer is silently retrying the request because it doesn’t receive an immediate header.

  • Could this be my Reverse Proxy (Nginx/Traefik) configuration? Does n8n require specific Keep-Alive settings for long-polling webhooks?

  • Is there a way to make the Webhook “Idempotent”? (i.e., force n8n to ignore a request if an identical one is already processing).

  • Should I switch architecture? For a 10s delay, should I switch the Webhook to “Respond Immediately” and push the result back to the frontend via a separate mechanism?

Any guidance on how to stop these

Capture d’écran 2025-11-22 à 03.45.48
Capture d’écran 2025-11-22 à 03.45.48600×1300 63.8 KB

Capture d’écran 2025-11-22 à 03.46.21
Capture d’écran 2025-11-22 à 03.46.212184×462 67 KB

auto-retries would be a lifesaver. I am ready for production, but this is a blocker.

Thanks a lot for your help! :folded_hands:

1 Like
2

You have just defined “the way” n8n supposed to work.

Before seeing any env or configuration, you’re almost certainly running into retry/duplicate calls coming from outside n8n rather than n8n “self‑restarting” the workflow.