TerraHome/nixos/kubeadm/scripts/reset-cluster-nodes.sh

#!/usr/bin/env bash
set -euo pipefail

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
INVENTORY_FILE="${1:-$SCRIPT_DIR/inventory.env}"

if [ ! -f "$INVENTORY_FILE" ]; then
  echo "Missing inventory file: $INVENTORY_FILE"
  echo "Copy $SCRIPT_DIR/inventory.example.env to $SCRIPT_DIR/inventory.env and edit node mappings."
  exit 1
fi

# shellcheck disable=SC1090
source "$INVENTORY_FILE"

SSH_USER="${SSH_USER:-micqdf}"
SSH_KEY_PATH="${SSH_KEY_PATH:-$HOME/.ssh/id_ed25519}"
SSH_OPTS="${SSH_OPTS:--o BatchMode=yes -o IdentitiesOnly=yes -o StrictHostKeyChecking=accept-new -i $SSH_KEY_PATH}"
SSH_USER_CANDIDATES="${SSH_USER_CANDIDATES:-$SSH_USER root}"

declare -A NODE_IPS=()

add_pair() {
  local pair="$1"
  local name="${pair%%=*}"
  local ip="${pair#*=}"

  if [ -z "$name" ] || [ -z "$ip" ] || [ "$name" = "$ip" ]; then
    echo "Invalid node pair '$pair' (expected name=ip)."
    exit 1
  fi

  NODE_IPS["$name"]="$ip"
}

if [ -n "${CONTROL_PLANES:-}" ]; then
  for pair in $CONTROL_PLANES; do
    add_pair "$pair"
  done
else
  while IFS= read -r var_name; do
    idx="${var_name#CP_}"
    add_pair "cp-$idx=${!var_name}"
  done < <(compgen -A variable | grep -E '^CP_[0-9]+$' | sort -V)
fi

if [ -n "${WORKERS:-}" ]; then
  for pair in $WORKERS; do
    add_pair "$pair"
  done
else
  while IFS= read -r var_name; do
    idx="${var_name#WK_}"
    add_pair "wk-$idx=${!var_name}"
  done < <(compgen -A variable | grep -E '^WK_[0-9]+$' | sort -V)
fi

if [ "${#NODE_IPS[@]}" -eq 0 ]; then
  echo "No nodes found in inventory."
  exit 1
fi

detect_ssh_user() {
  local probe_ip="$1"
  local candidate

  for candidate in $SSH_USER_CANDIDATES; do
    if ssh $SSH_OPTS "$candidate@$probe_ip" "true" >/dev/null 2>&1; then
      ACTIVE_SSH_USER="$candidate"
      echo "==> Using SSH user '$ACTIVE_SSH_USER'"
      return 0
    fi
  done

  echo "Unable to authenticate to $probe_ip with candidates: $SSH_USER_CANDIDATES"
  return 1
}

mkdir -p "$HOME/.ssh"
chmod 700 "$HOME/.ssh"
touch "$HOME/.ssh/known_hosts"
chmod 600 "$HOME/.ssh/known_hosts"
for node_name in "${!NODE_IPS[@]}"; do
  ssh-keygen -R "${NODE_IPS[$node_name]}" >/dev/null 2>&1 || true
  ssh-keyscan -H "${NODE_IPS[$node_name]}" >> "$HOME/.ssh/known_hosts" 2>/dev/null || true
done

reset_node() {
  local node_name="$1"
  local node_ip="$2"
  echo "==> Resetting $node_name ($node_ip)"
  ssh $SSH_OPTS "$ACTIVE_SSH_USER@$node_ip" "sudo kubeadm reset -f && sudo systemctl stop kubelet && sudo rm -rf /etc/kubernetes /var/lib/etcd /var/lib/cni /etc/cni/net.d"
}

FIRST_NODE_IP="${NODE_IPS[$(printf '%s\n' "${!NODE_IPS[@]}" | sort -V | head -n1)]}"
ACTIVE_SSH_USER="$SSH_USER"
detect_ssh_user "$FIRST_NODE_IP"

while IFS= read -r node_name; do
  reset_node "$node_name" "${NODE_IPS[$node_name]}"
done < <(printf '%s\n' "${!NODE_IPS[@]}" | sort -V)

echo "Cluster components reset on all listed nodes."