I am currently investigating an issue where there is some kind of networking change in the beginning of the pod lifecycle.
Workloads that immediately start sending traffic can see dropped requests. There is high pod churn.
We investigated kube-proxy, aws-cni eni setup, deactivated aws-network-policy-agent Network Policy setup, watched ip addr, iptables and were unable to correlate a root cause.
A sleep 1s in an init container does solve this issue.
Is there any component that are also of interest to investigate?
Cluster information:
Kubernetes version: 1.30
Cloud being used: aws eks
Installation method: terraform
Host OS: AL2
CNI and version: amazon-k8s-cni-init:v1.18.2-eksbuild.1