controller join not working with AWS NLB about k0s HOT 2 OPEN

After the log line "initialized join client successfully", the second controller attempts to make an HTTP request to obtain certificates and key material from the first controller. My educated guess is that this request is blocking on network I/O. The load balancer is probably not responding about 50% of the time because it is balancing the load between the two controllers. (I assume the load balancer is using a round-robin or random load balancing strategy). When the load balancer tries to connect to the secondary controller, it fails because the secondary controller cannot respond to a join request from itself. It's not even listening on the join port yet. So I think the LB is trying to reach the secondary controller, keeping the HTTP request open and not responding yet. K0s retries join requests up to ten times. However, it doesn't define any timeouts, so it probably gets stuck on the first request.

Can you maybe check the configuration of your load balancer? Does it have some backend health checks defined? What are the frontend and backend timeouts?

from k0s.

Thankyou for taking the time to respond. My hunch is that the issue lies with the load balancer. Yesterday I tore down the LB and backend hosts and recreated them. Since then, I have done many reset/apply actions and I haven't seen the issue occur. This time the 'client IP preservation' setting is disabled. I had previously disabled it, but still saw the issue occur (or so I thought), so I'm confused why it seems to be consistently working now...

However, it doesn't define any timeouts, so it probably gets stuck on the first request.

Would you be willing to add timeouts for read/write? I think this would be good in terms of best practice and to aid in debugging these kinds of unusual scenarios.

from k0s.