wild-cloud/wild-cli/cmd/wild/setup/cluster.go

package setup

import (
	"context"
	"fmt"
	"os"
	"path/filepath"

	"github.com/spf13/cobra"

	"github.com/wild-cloud/wild-cli/internal/config"
	"github.com/wild-cloud/wild-cli/internal/environment"
	"github.com/wild-cloud/wild-cli/internal/external"
	"github.com/wild-cloud/wild-cli/internal/output"
)

var (
	skipInstaller bool
	skipHardware  bool
)

func newClusterCommand() *cobra.Command {
	cmd := &cobra.Command{
		Use:   "cluster",
		Short: "Set up Kubernetes cluster",
		Long: `Set up the Kubernetes cluster infrastructure using Talos Linux.

This command configures Talos Linux nodes and bootstraps the Kubernetes cluster.

Examples:
  wild setup cluster
  wild setup cluster --skip-installer
  wild setup cluster --skip-hardware`,
		RunE: runCluster,
	}

	cmd.Flags().BoolVar(&skipInstaller, "skip-installer", false, "skip installer image generation")
	cmd.Flags().BoolVar(&skipHardware, "skip-hardware", false, "skip node hardware detection")

	return cmd
}

func runCluster(cmd *cobra.Command, args []string) error {
	output.Header("Wild Cloud Cluster Setup")

	// Initialize environment
	env := environment.New()
	if err := env.RequiresProject(); err != nil {
		return err
	}

	// Check external tools
	toolManager := external.NewManager()
	if err := toolManager.CheckTools(cmd.Context(), []string{"talosctl"}); err != nil {
		return fmt.Errorf("required tools not available: %w", err)
	}

	// Load configuration
	configMgr := config.NewManager(env.ConfigPath(), env.SecretsPath())

	// Get cluster configuration
	clusterName, err := getConfigString(configMgr, "cluster.name")
	if err != nil {
		return fmt.Errorf("cluster name not configured: %w", err)
	}

	vip, err := getConfigString(configMgr, "cluster.vip")
	if err != nil {
		return fmt.Errorf("cluster VIP not configured: %w", err)
	}

	output.Info("Cluster: " + clusterName)
	output.Info("VIP: " + vip)

	// Phase 1: Generate Talos configuration
	output.Info("\n=== Phase 1: Generating Talos Configuration ===")

	configDir := filepath.Join(env.WildCloudDir(), "talos")
	if err := os.MkdirAll(configDir, 0755); err != nil {
		return fmt.Errorf("creating config directory: %w", err)
	}

	talosctl := toolManager.Talosctl()
	clusterEndpoint := "https://" + vip + ":6443"

	if err := talosctl.GenerateConfig(cmd.Context(), clusterName, clusterEndpoint, configDir); err != nil {
		return fmt.Errorf("generating talos config: %w", err)
	}

	output.Success("Talos configuration generated")

	// Phase 2: Node configuration
	if !skipHardware {
		output.Info("\n=== Phase 2: Detecting Nodes ===")
		if err := detectAndConfigureNodes(cmd.Context(), configMgr, talosctl, configDir); err != nil {
			return fmt.Errorf("configuring nodes: %w", err)
		}
	} else {
		output.Info("Skipping node hardware detection")
	}

	// Phase 3: Bootstrap cluster
	output.Info("\n=== Phase 3: Bootstrapping Cluster ===")
	if err := bootstrapCluster(cmd.Context(), configMgr, talosctl, configDir); err != nil {
		return fmt.Errorf("bootstrapping cluster: %w", err)
	}

	output.Success("Cluster setup completed successfully!")
	output.Info("")
	output.Info("Next steps:")
	output.Info("  wild setup services           # Install cluster services")
	output.Info("  kubectl get nodes              # Verify cluster")

	return nil
}

// detectAndConfigureNodes detects and configures cluster nodes
func detectAndConfigureNodes(ctx context.Context, configMgr *config.Manager, talosctl *external.TalosctlTool, configDir string) error {
	// Get nodes from configuration
	nodesConfig, err := configMgr.Get("cluster.nodes")
	if err != nil || nodesConfig == nil {
		output.Warning("No nodes configured")
		output.Info("Add nodes to config: wild config set cluster.nodes '[{\"ip\": \"192.168.1.10\", \"role\": \"controlplane\"}]'")
		return nil
	}

	nodes, ok := nodesConfig.([]interface{})
	if !ok {
		return fmt.Errorf("invalid nodes configuration")
	}

	if len(nodes) == 0 {
		output.Warning("No nodes configured")
		return nil
	}

	output.Info(fmt.Sprintf("Found %d nodes in configuration", len(nodes)))

	// Configure each node
	for i, nodeConfig := range nodes {
		nodeMap, ok := nodeConfig.(map[string]interface{})
		if !ok {
			output.Warning(fmt.Sprintf("Invalid node %d configuration", i))
			continue
		}

		nodeIP, exists := nodeMap["ip"]
		if !exists {
			output.Warning(fmt.Sprintf("Node %d missing IP address", i))
			continue
		}

		nodeRole, exists := nodeMap["role"]
		if !exists {
			nodeRole = "worker"
		}

		output.Info(fmt.Sprintf("Configuring node %s (%s)", nodeIP, nodeRole))

		// Apply configuration to node
		var configFile string
		if nodeRole == "controlplane" {
			configFile = filepath.Join(configDir, "controlplane.yaml")
		} else {
			configFile = filepath.Join(configDir, "worker.yaml")
		}

		talosctl.SetEndpoints([]string{fmt.Sprintf("%v", nodeIP)})
		if err := talosctl.ApplyConfig(ctx, configFile, true); err != nil {
			output.Warning(fmt.Sprintf("Failed to configure node %s: %v", nodeIP, err))
		} else {
			output.Success(fmt.Sprintf("Node %s configured", nodeIP))
		}
	}

	return nil
}

// bootstrapCluster bootstraps the Kubernetes cluster
func bootstrapCluster(ctx context.Context, configMgr *config.Manager, talosctl *external.TalosctlTool, configDir string) error {
	// Get first controlplane node
	nodesConfig, err := configMgr.Get("cluster.nodes")
	if err != nil || nodesConfig == nil {
		return fmt.Errorf("no nodes configured")
	}

	nodes, ok := nodesConfig.([]interface{})
	if !ok || len(nodes) == 0 {
		return fmt.Errorf("invalid nodes configuration")
	}

	// Find first controlplane node
	var bootstrapNode string
	for _, nodeConfig := range nodes {
		nodeMap, ok := nodeConfig.(map[string]interface{})
		if !ok {
			continue
		}

		nodeIP, exists := nodeMap["ip"]
		if !exists {
			continue
		}

		nodeRole, exists := nodeMap["role"]
		if exists && nodeRole == "controlplane" {
			bootstrapNode = fmt.Sprintf("%v", nodeIP)
			break
		}
	}

	if bootstrapNode == "" {
		return fmt.Errorf("no controlplane node found")
	}

	output.Info("Bootstrap node: " + bootstrapNode)

	// Set talosconfig
	talosconfig := filepath.Join(configDir, "talosconfig")
	talosctl.SetTalosconfig(talosconfig)
	talosctl.SetEndpoints([]string{bootstrapNode})
	talosctl.SetNodes([]string{bootstrapNode})

	// Bootstrap cluster
	if err := talosctl.Bootstrap(ctx); err != nil {
		return fmt.Errorf("bootstrapping cluster: %w", err)
	}

	output.Success("Cluster bootstrapped")

	// Generate kubeconfig
	output.Info("Generating kubeconfig...")
	kubeconfigPath := filepath.Join(configDir, "kubeconfig")
	if err := talosctl.Kubeconfig(ctx, kubeconfigPath, true); err != nil {
		output.Warning("Failed to generate kubeconfig: " + err.Error())
	} else {
		output.Success("Kubeconfig generated: " + kubeconfigPath)
		output.Info("Set KUBECONFIG=" + kubeconfigPath)
	}

	return nil
}

// getConfigString gets a string value from config with validation
func getConfigString(configMgr *config.Manager, path string) (string, error) {
	value, err := configMgr.Get(path)
	if err != nil {
		return "", err
	}

	if value == nil {
		return "", fmt.Errorf("config value '%s' not set", path)
	}

	strValue, ok := value.(string)
	if !ok {
		return "", fmt.Errorf("config value '%s' is not a string", path)
	}

	if strValue == "" {
		return "", fmt.Errorf("config value '%s' is empty", path)
	}

	return strValue, nil
}