#!/sbin/openrc-run
# Konnectivity server (apiserver-network-proxy) — the apiserver egress proxy for
# the Mulga EKS control plane. The kube-apiserver's `cluster` egress selector
# (egress-selector-config.yaml) routes pod/kubelet/webhook traffic to this
# server's UDS; konnectivity-agents on the workers dial in over :8132 and the
# egress rides back down the agent-initiated reverse tunnel. Server-role only;
# enabled by eks-node-role. Runs as root to bind the UDS the apiserver reads.

description="Konnectivity server (Mulga EKS apiserver egress proxy)"

KONN_DIR=/var/lib/spinifex-eks/konnectivity
KONN_UDS=/run/konnectivity/konnectivity-server.socket
ENVFILE=/etc/spinifex-eks/first-boot.env
K3S_CA=/var/lib/rancher/k3s/server/tls/server-ca.crt
K3S_CA_KEY=/var/lib/rancher/k3s/server/tls/server-ca.key
KUBECONFIG_FILE=/etc/rancher/k3s/k3s.yaml

# server-count is the apiserver replica count: each agent holds a tunnel to every
# replica, so egress from any apiserver always resolves (the HA-correct fan-out).
# Seeded by the launcher in first-boot.env; default 1 for a single control plane.
_server_count=1
if [ -f "${ENVFILE}" ]; then
    _sc=$(awk -F= '/^EKS_KONNECTIVITY_SERVER_COUNT=/{print $2}' "${ENVFILE}")
    [ -n "${_sc}" ] && _server_count="${_sc}"
fi

command="/usr/local/bin/konnectivity-server"
command_args="--logtostderr=true
 --uds-name=${KONN_UDS}
 --delete-existing-uds-file
 --cluster-cert=${KONN_DIR}/tls.crt
 --cluster-key=${KONN_DIR}/tls.key
 --server-count=${_server_count}
 --mode=grpc
 --server-port=0
 --agent-port=8132
 --admin-port=8093
 --health-port=8092
 --kubeconfig=${KUBECONFIG_FILE}
 --authentication-audience=system:konnectivity-server
 --agent-namespace=kube-system
 --agent-service-account=konnectivity-agent
 --proxy-strategies=destHost,default"
pidfile="/run/konnectivity-server.pid"

# Supervise: a transient exit must self-heal, or apiserver egress (webhooks,
# exec, logs) stays broken until the next boot and the cluster wedges.
supervisor="supervise-daemon"
respawn_delay=5
respawn_max=0
output_log="/var/log/konnectivity-server.log"
error_log="/var/log/konnectivity-server.log"

depend() {
    need net
    after k3s
}

start_pre() {
    mkdir -p /run/konnectivity "${KONN_DIR}"

    # Wait for k3s to write the server CA (+ key) and admin kubeconfig: the CA
    # signs the serving cert agents validate, the kubeconfig is the TokenReview
    # client for agent SA-token auth. Bounded; without them konnectivity is inert.
    i=0
    while [ "${i}" -lt 120 ]; do
        if [ -r "${K3S_CA}" ] && [ -r "${K3S_CA_KEY}" ] && [ -r "${KUBECONFIG_FILE}" ]; then
            break
        fi
        i=$((i + 2))
        sleep 2
    done
    if [ ! -r "${K3S_CA}" ] || [ ! -r "${K3S_CA_KEY}" ] || [ ! -r "${KUBECONFIG_FILE}" ]; then
        eerror "k3s server CA / admin kubeconfig not present after ${i}s"
        return 1
    fi

    # Mint (or reuse) the serving cert from the k3s server CA, SANed to the NLB
    # endpoints agents dial. That CA is the in-pod ca.crt the agent already trusts,
    # so no separate CA distribution is needed.
    sans=""
    if [ -f "${ENVFILE}" ]; then
        sans=$(awk -F= '/^EKS_KONNECTIVITY_SANS=/{print $2}' "${ENVFILE}")
    fi
    if [ -z "${sans}" ]; then
        eerror "EKS_KONNECTIVITY_SANS empty; agents would fail server cert validation"
        return 1
    fi
    if ! eks-konnectivity-cert -dir "${KONN_DIR}" -ca-cert "${K3S_CA}" \
            -ca-key "${K3S_CA_KEY}" -cn konnectivity-server -sans "${sans}" >/dev/null; then
        eerror "failed to mint konnectivity serving cert"
        return 1
    fi
}