* feat: add domain virtual responses with a compact editor and bounded edge matching
* fix: keep traffic working during rolling edge script updates
* refactor: centralize edge management transport
* fix(scaling): refuse cgroup v1 hosts and surface k3s crash causes
K3s v1.36's kubelet refuses to run on cgroup v1 hosts, so a server that
passed inspection could still crash-loop after install. The readiness
check then reported only "connection refused" against the API port,
hiding the real failure under etcd log noise.
- inspect() now fails before install when /sys/fs/cgroup/cgroup.controllers
is missing, telling the user to boot with systemd.unified_cgroup_hierarchy=1.
- readiness() now reports the unit state, restart count and the last fatal
journal lines whenever the service is not active or the API is unreachable.
* fix(scaling): make the local server the first cluster control
The first control server is the cluster's API, build and Edge gateway,
but roles were assigned purely by sorted server ID. On a fleet where only
"This Server" has a public address, that server could land as a worker,
leaving the gateway on a private-only node.
Order the local server first; other members keep their sorted ID order.
* fix(scaling): dial worker kubelets directly instead of via the agent tunnel
With the default egress-selector-mode (agent), the K3s server reaches a
worker's kubelet through the agent tunnel, and the agent dials the kubelet
on loopback. OpenShip pins the kubelet to the private address, so that
dial is refused and kubectl exec/logs on workers fail with
"error dialing backend: EOF". Cluster setup then fails at the routing
check. Control servers were unaffected because the API server dials them
directly.
Set egress-selector-mode to disabled on servers so the API server dials
every kubelet at its private address, which the peer firewall rules
already allow.
---------
Co-authored-by: Hydra <abdelhamedmo59@gmail.com>