High CPU usage and unresponsive dashboard after running long Loop workflow

[Polonco]

Bug Description

I am self-hosting n8n using Docker on a DigitalOcean droplet with the following specifications:

Regular Intel droplet (1 vCPU, 1 GB RAM)
OS: Ubuntu 22.04

I encountered an issue when running a workflow designed to loop approximately 3,000 times.

Workflow Details:

1. The workflow sends a photo (fetched from a URL) and a message to a Telegram user.
2. It waits 10 seconds, then repeats the process for the next Telegram user.
3. The loop continues for all 3,000 Telegram user IDs.

Observed Behavior:

• After approximately 1,000 iterations (450 minutes of runtime), the CPU usage and server load increased significantly.
• The n8n dashboard became unresponsive.
• Other workflows also became extremely slow to execute.
• The workflow started at 15:30 and was manually stopped at around 23:00 due to these issues.

Workflow start at 15:30. I manually stopped it at about 23:00

To Reproduce

1. Extract 3,000 Telegram user IDs.
2. Create a workflow with a loop that:
3. Sends a Telegram message with a photo (hosted on a website and fetched via URL).
4. Waits 10 seconds.
5. Sends the same content to the next user.
6. Monitor the server performance during the workflow execution.

Expected behavior

Expected behavior

• The server load and CPU usage should remain stable even during long-running workflows.
• The n8n dashboard and other workflows should remain responsive throughout.

Actual Behavior:

• Server load and CPU usage become excessively high after prolonged execution of the workflow.
• n8n dashboard and other workflows become unresponsive.

Operating System

Ubuntu Linux 22.04

n8n Version

1.72.1

Node.js Version

20.18.0

Database

SQLite (default)

Execution mode

main (default)

[Joffcom]

Hey @Polonco,

We have created an internal ticket to look into this which we will be tracking as "N8N-8001"

[Joffcom]

Hey @Polonco,

It may be better to optimise your workflow so you are not keeping all of that in the workflow while it is running. I would move the 3 telegram nodes to a sub workflow and just return a simple bool that way the response from those nodes and the processing will be cleaned up when the sub workflow finishes instead of being added to the main workflow.