wild-cloud/bin/wild-cluster-node-up

#!/bin/bash

set -e
set -o pipefail

# Usage function
usage() {
    echo "Usage: wild-cluster-node-up <node-name> [options]"
    echo ""
    echo "Apply Talos machine configuration to a registered node."
    echo ""
    echo "Arguments:"
    echo "  node-name           Name of the registered node"
    echo ""
    echo "Options:"
    echo "  -i, --insecure      Apply configuration in insecure mode (for maintenance mode nodes)"
    echo "  --force             Force regeneration of final config even if it exists"
    echo "  --dry-run           Show the command that would be executed without running it"
    echo "  -h, --help          Show this help message"
    echo ""
    echo "Examples:"
    echo "  wild-cluster-node-up control-1"
    echo "  wild-cluster-node-up worker-1 --insecure"
    echo "  wild-cluster-node-up worker-2 --skip-patch"
    echo "  wild-cluster-node-up control-2 --force"
    echo "  wild-cluster-node-up control-1 --dry-run"
    echo ""
    echo "This script will:"
    echo "  - Verify the node is registered in config.yaml"
    echo "  - Generate final machine configuration if needed"
    echo "  - Apply the configuration using talosctl apply-config"
    echo "  - Use insecure mode for nodes in maintenance mode"
    echo ""
    echo "Requirements:"
    echo "  - Must be run from a wild-cloud directory"
    echo "  - Node must be registered (hardware detected) first"
    echo "  - Base cluster configuration and patch file must exist for the node"
}

# Parse arguments
NODE_NAME=""
INSECURE_MODE=false
DRY_RUN=false
SKIP_PATCH=false
FORCE_REGENERATE=false

while [[ $# -gt 0 ]]; do
    case $1 in
        -i|--insecure)
            INSECURE_MODE=true
            shift
            ;;
        --force)
            FORCE_REGENERATE=true
            shift
            ;;
        --dry-run)
            DRY_RUN=true
            shift
            ;;
        -h|--help)
            usage
            exit 0
            ;;
        -*)
            echo "Unknown option $1"
            usage
            exit 1
            ;;
        *)
            if [ -z "$NODE_NAME" ]; then
                NODE_NAME="$1"
            else
                echo "Unexpected argument: $1"
                usage
                exit 1
            fi
            shift
            ;;
    esac
done

# Check if node name was provided
if [ -z "$NODE_NAME" ]; then
    echo "Error: Node name is required"
    usage
    exit 1
fi

# Initialize Wild Cloud environment
if [ -z "${WC_ROOT}" ]; then
    print "WC_ROOT is not set."
    exit 1
else
    source "${WC_ROOT}/scripts/common.sh"
    init_wild_env
fi

print_header "Talos node configuration"

# Check if the specified node is registered
NODE_INTERFACE=$(yq eval ".cluster.nodes.active.\"${NODE_NAME}\".interface" "${WC_HOME}/config.yaml" 2>/dev/null)
NODE_DISK=$(yq eval ".cluster.nodes.active.\"${NODE_NAME}\".disk" "${WC_HOME}/config.yaml" 2>/dev/null)
NODE_ROLE=$(yq eval ".cluster.nodes.active.\"${NODE_NAME}\".role" "${WC_HOME}/config.yaml" 2>/dev/null)
NODE_CURRENT_IP=$(yq eval ".cluster.nodes.active.\"${NODE_NAME}\".currentIp" "${WC_HOME}/config.yaml" 2>/dev/null)
MAINTENANCE_IP=$(yq eval ".cluster.nodes.active.\"${NODE_NAME}\".maintenanceIp" "${WC_HOME}/config.yaml" 2>/dev/null)

if [ -z "$NODE_INTERFACE" ] || [ "$NODE_INTERFACE" = "null" ]; then
    print_error "Node $NODE_NAME is not registered in config.yaml"
    print_info "Please register the node first by running:"
    print_info "Or run 'wild-setup-cluster' to register nodes interactively"
    exit 1
fi

# Get current IP for the node
if [ -z "$NODE_CURRENT_IP" ] || [ "$NODE_CURRENT_IP" = "null" ]; then
    print_error "Node $NODE_NAME has no current IP address set"
    exit 1
fi

# Determine node type
if [ "$NODE_ROLE" = "controlplane" ]; then
    NODE_TYPE="control plane"
else
    NODE_TYPE="worker"
fi

# Determine the target IP for applying configuration
if [ -n "$MAINTENANCE_IP" ] && [ "$MAINTENANCE_IP" != "null" ]; then
    TARGET_IP="$MAINTENANCE_IP"
    print_info "Applying configuration to $NODE_TYPE node: $NODE_NAME ($NODE_CURRENT_IP) via maintenance IP: $MAINTENANCE_IP"
    # Auto-enable insecure mode when using maintenance IP (unless explicitly overridden)
    if [ "$INSECURE_MODE" = false ]; then
        INSECURE_MODE=true
        print_info "Auto-enabling insecure mode for maintenance IP"
    fi
else
    TARGET_IP="$NODE_CURRENT_IP"
    print_info "Applying configuration to $NODE_TYPE node: $NODE_NAME ($NODE_CURRENT_IP)"
fi

print_info "Node details:"
print_info "  - Name: $NODE_NAME"
print_info "  - Current IP: $NODE_CURRENT_IP"
print_info "  - Interface: $NODE_INTERFACE"
print_info "  - Disk: $NODE_DISK"
print_info "  - Type: $NODE_TYPE"
if [ -n "$MAINTENANCE_IP" ] && [ "$MAINTENANCE_IP" != "null" ]; then
    print_info "  - Maintenance IP: $MAINTENANCE_IP"
fi

# Check if machine config exists, generate if needed
NODE_SETUP_DIR="${WC_HOME}/setup/cluster-nodes"
CONFIG_FILE="${NODE_SETUP_DIR}/final/${NODE_NAME}.yaml"
PATCH_FILE="${NODE_SETUP_DIR}/patch/${NODE_NAME}.yaml"

# Check if patch file exists
if [ ! -f "$PATCH_FILE" ]; then
    wild-cluster-node-patch-generate "$NODE_NAME"
fi

# Determine base config file
if [ "$NODE_ROLE" = "controlplane" ]; then
    BASE_CONFIG="${NODE_SETUP_DIR}/generated/controlplane.yaml"
else
    BASE_CONFIG="${NODE_SETUP_DIR}/generated/worker.yaml"
fi

# Check if base config exists
if [ ! -f "$BASE_CONFIG" ]; then
    print_error "Base configuration not found: $BASE_CONFIG"
    print_info "Generate base cluster configuration first:"
    print_info "  wild-cluster-config-generate"
    exit 1
fi

# Check if we should skip regeneration
if [ ! -f "$CONFIG_FILE" ] || [ "$FORCE_REGENERATE" = true ]; then
    # Need to generate/regenerate the final config
    if [ "$FORCE_REGENERATE" = true ]; then
        print_info "Force regeneration requested: regenerating machine configuration..."
    else
        print_info "Machine configuration not found: $CONFIG_FILE"
        print_info "Generating final machine configuration..."
    fi
        
    # Create final config directory if it doesn't exist
    mkdir -p "${NODE_SETUP_DIR}/final"
    
    # Generate final machine config
    print_info "Generating final machine configuration from patch..."
    talosctl machineconfig patch "$BASE_CONFIG" --patch @"$PATCH_FILE" -o "$CONFIG_FILE"
    print_success "Generated machine configuration: $CONFIG_FILE"
else
    print_success "Found existing machine configuration: $CONFIG_FILE"
fi

# Build talosctl command
TALOSCTL_CMD="talosctl apply-config"

if [ "$INSECURE_MODE" = true ]; then
    TALOSCTL_CMD="$TALOSCTL_CMD --insecure"
    print_info "Using insecure mode (for maintenance mode nodes)"
fi

TALOSCTL_CMD="$TALOSCTL_CMD --nodes $TARGET_IP --file $CONFIG_FILE"

# Show the command
echo ""
print_info "Command to execute:"
echo "  $TALOSCTL_CMD"
echo ""

if [ "$DRY_RUN" = true ]; then
    print_info "Dry run mode - command shown above but not executed"
    exit 0
fi

# Apply the configuration
print_info "Applying machine configuration..."
echo ""

if eval "$TALOSCTL_CMD"; then
    print_success "Machine configuration applied successfully!"
    
    # Update talosctl context to this node
    print_info "Updating talosctl context..."
    talosctl config node "$NODE_CURRENT_IP"
    print_success "Updated talosctl context to node $NODE_NAME ($NODE_CURRENT_IP)"
    echo ""
    
    if [ "$NODE_ROLE" = "controlplane" ]; then
        print_info "Next steps for control plane node:"
        echo "  1. Wait for the node to reboot and come up with the new configuration"
        echo "  2. If this is your first control plane node, bootstrap it:"
        echo "     talosctl bootstrap --nodes $NODE_CURRENT_IP"
        echo "  3. Get kubeconfig when cluster is ready:"
        echo "     talosctl kubeconfig"
    else
        print_info "Next steps for worker node:"
        echo "  1. Wait for the node to reboot and come up with the new configuration"
        echo "  2. Node will join the cluster automatically"
        echo "  3. Verify the node appears in the cluster:"
        echo "     kubectl get nodes"
    fi
    
    echo ""
    print_info "Monitor node status with:"
    echo "  talosctl --nodes $NODE_CURRENT_IP dmesg"
    echo "  talosctl --nodes $NODE_CURRENT_IP get members"
    
else
    print_error "Failed to apply machine configuration"
    echo ""
    print_info "Troubleshooting tips:"
    if [ -n "$MAINTENANCE_IP" ] && [ "$MAINTENANCE_IP" != "null" ]; then
        echo "  - Ensure the node is accessible at maintenance IP $MAINTENANCE_IP"
    else
        echo "  - Ensure the node is accessible at $NODE_CURRENT_IP"
    fi
    echo "  - For nodes in maintenance mode, use --insecure flag"
    echo "  - Check network connectivity and firewall settings"
    echo "  - Verify the machine configuration file is valid"
    exit 1
fi

print_success "Node configuration completed!"