Track  AWS support case to understand the high target reset count metric on k8s-apiserver on default IPI deployment.

That issue was perceived when investigating other issue with NLB, I am assuming that the reset connections comes from health check from balancer nodes for two reasons:

• the public NLB is idle, and the reset connections are linear
• when I changed the target to TCP checks, the reset metric dropped. So the HTTP and HTTPS could be the cause of that, for that reason we need to understand why the TCP communication between the balancer and targets (apiserver) is not closing correctly doing simple HTTP requests

As far we know, it's not impacting nothing on apiserver more than the "high Target Reset Count metric" of TCP reset connections that comes from health checks. Despite of it's not generating any perceived issue the root cause should tracked and documented to have a complete understanding of our solution on AWS.

References:

• AWS support case (openshift-dev account)
• Internal document of this research