Queue mode uses the main option and is able to handle multiple workflow executions at once because it only really has one job. You can find a bit more information about this here: Configuring queue mode | n8n Docs
Okay, understood, I know with own mode if a process crashed it wouldn’t have affected others, but now in queue it perhaps would crash all workflows running on that worker.
I have had a thought of perhaps mitigating the impact by reducing the concurrency on the workers, and just scale up when needed.
Is there some metric that n8n can report or some example you know of where I can scale (I’m using horizontal pod autoscaler) based on ongoing executions not just the pods resource usage.
That would help a lot with this issue.
It is probably worth noting that queue mode works in a different way to a main instance so generally you would never really try to use own mode with it as it is designed in a different way. When to scale would depend on what your workflows are doing so I would probably start with 2 worker nodes and run a bunch of workflows to see which metric works for your environment. It is the sort of thing where I suspect there is no one size fits all and it depends on what you are doing.