48 lines
1.6 KiB
Bash
48 lines
1.6 KiB
Bash
|
#!/usr/bin/env bash
|
||
|
|
set -euo pipefail
|
||
|
|
|
||
|
|
host_root="/host"
|
||
|
|
unit="k3s-agent"
|
||
|
|
unit_file="${host_root}/etc/systemd/system/${unit}.service"
|
||
|
|
config_dir="${host_root}/etc/rancher/k3s/config.yaml.d"
|
||
|
|
config_file="${config_dir}/90-atlas-rpi-reservations.yaml"
|
||
|
|
|
||
|
|
if [ ! -f "${unit_file}" ]; then
|
||
|
|
echo "k3s-agent unit not found; this guardrail only manages worker agents"
|
||
|
|
sleep infinity
|
||
|
|
fi
|
||
|
|
|
||
|
|
tmp_file="$(mktemp)"
|
||
|
|
cat > "${tmp_file}" <<'EOF'
|
||
|
|
# Managed by Flux via services/maintenance/scripts/rpi_resource_reservation.sh.
|
||
|
|
# Keep RPi workers below saturation so kubelet and the OS keep enough headroom
|
||
|
|
# to evict or recover before the board wedges.
|
||
|
|
kubelet-arg+:
|
||
|
|
- "system-reserved=cpu=250m,memory=384Mi,ephemeral-storage=1Gi"
|
||
|
|
- "kube-reserved=cpu=150m,memory=256Mi,ephemeral-storage=1Gi"
|
||
|
|
- "eviction-hard=memory.available<512Mi,nodefs.available<10%,imagefs.available<10%"
|
||
|
|
- "eviction-soft=memory.available<768Mi,nodefs.available<15%,imagefs.available<15%"
|
||
|
|
- "eviction-soft-grace-period=memory.available=1m,nodefs.available=2m,imagefs.available=2m"
|
||
|
|
- "eviction-max-pod-grace-period=60"
|
||
|
|
EOF
|
||
|
|
|
||
|
|
changed=0
|
||
|
|
if [ ! -f "${config_file}" ] || ! cmp -s "${tmp_file}" "${config_file}"; then
|
||
|
|
mkdir -p "${config_dir}"
|
||
|
|
install -m 0644 "${tmp_file}" "${config_file}"
|
||
|
|
changed=1
|
||
|
|
fi
|
||
|
|
rm -f "${tmp_file}"
|
||
|
|
|
||
|
|
if [ "${changed}" -eq 1 ]; then
|
||
|
|
delay="$(( (RANDOM % 420) + 30 ))"
|
||
|
|
echo "updated ${config_file}; restarting ${unit} after ${delay}s"
|
||
|
|
sleep "${delay}"
|
||
|
|
chroot "${host_root}" /bin/systemctl daemon-reload
|
||
|
|
chroot "${host_root}" /bin/systemctl restart "${unit}"
|
||
|
|
else
|
||
|
|
echo "${config_file} already up to date"
|
||
|
|
fi
|
||
|
|
|
||
|
|
sleep infinity
|