Everything is new
- kiac-lb replaces MetalLB: a tiny controller inside the control-plane VM assigns endpoint-local node IPs to LoadBalancer Services in ~2s, shares IPs across services on disjoint ports, and self-heals after node restarts. No pods, no webhooks, no ARP speakers.
--distro k3s: rancher/k3s as PID 1 per node VM. sqlite datastore, bundled servicelb/local-path/metrics-server, kindnet networking. Two-node cluster in ~30s at ~3.7GB host RSS.- Cilium on a Mac:
kiac create cluster --cni cilium --kernel fullboots nodes on a published custom kernel (VXLAN, Geneve, br_netfilter, nf_tables, BPF JIT/BTF, WireGuard) and installs Cilium's eBPF datapath. Bonus: the vxlan overlay rides vmnet's fast path at ~285MB/s cross-node, ~1GB/s Mac-to-pod. kiac resume: clusters survive host reboots. Restarts VMs and heals the control-plane address everywhere it is pinned (apiserver cert, kubeconfigs, kube-proxy).- Observability + Gateway API on every distro:
--observability(Prometheus + Grafana with provisioned dashboards) and--gateway(Gateway API + Traefik) verified on kubeadm, k3s, and Cilium clusters. - Node chaos (
kiac stop/start node), declarative--config, shell completions,doctor --fix,get clusters -o json, and a dashboard with an embedded kubectl console (kiac ui). - Faster creates (parallel boot/join/addons), 2G worker default with
--cp-memory.
Docs: https://saiyam1814.github.io/kiac/ · Kernel: kernel-v6.12.28-full release
Install: brew install saiyam1814/tap/kiac