Hello,
I am looking a project ‘integrations’ implementation for a client who wants to get rid of a bunch of home grown applications and replace with something cool, modern, and easier to maintain solutions like n8n.
I have a few questions around the setup and architecture
Do you have any performance and scalability numbers n8n? At what point does the system failover
What are deployment architecture considerations? Is the system just database bound in terms of performance bottleneck? Can I just stick this on a Kubernetes auto-scaling pod and would this work fine or is there some big database bottleneck that I should be aware of.
We want to install some monitoring tool like datadog so that we can forward the application logs to collect useful information. Where can we find useful logs for n8n
Very sorry @Ali_Razat this topic must have slipped through.
No sorry. Sadly nothing. Would also be hard to do as the main factors are the hardware it runs on and the workflows it runs (and they can take literally anything between a few ms with almost no CPU and memory required or run days which high memory & CPU usage.
Scaling n8n is currently not that easy. You can find some information about it if you search for “scale/scaling” in the forum like this topic. But generally does n8n not scale properly yet. There is currently an open PR that adds a job-queue to n8n (like mentioned in the other topic). Is still in alpha but you can find it here if you want to test and give feedback. The two main bottlenecks would then be the DB used and Redis (which gets used by the job-queue).
That is something else where n8n is currently more than lacking. But It is also planned to improve that very soon.
Hi Jan, appreciate your open statements on the limits of n8n so far! Would you mind giving a short update on the points you’ve mentioned? I see the PR for a Redis queue has been merged, how did that impact scalability from your perspective?
Yes, that is correct. You can find the documentation here:
If you set up n8n as described above, will you be able to separately scale workers and webhooks-services according to your need. That will enable you to have a few million executions a day without a problem.
