Merge pull request 'fix: handle ANSI Terraform empty state' (#18) from fix/terraform-no-state-ansi into main

Reviewed-on: #18

internal/bootstrap/bootstrap.go

@@ -110,19 +110,18 @@ func listTerraformStateResources(terraformDir string, environment []string) ([]s
 	return strings.Fields(string(state)), nil
 }
 
+var ansiEscapeSequence = regexp.MustCompile(`\x1b\[[0-?]*[ -/]*[@-~]`)
+
 func terraformNoStateFile(stderr []byte) bool {
-	for _, line := range strings.Split(string(stderr), "\n") {
+	return strings.Contains(strings.ToLower(ansiEscapeSequence.ReplaceAllString(string(stderr), "")), "no state file")
-		if strings.TrimSpace(line) == "No state file was found!" {
-			return true
-		}
-	}
-	return false
 }
 
 var destroyTalosVMs = func(terraformDir string, environment []string) error {
 	return utils.RunCommandInDirEnv(terraformDir, environment, "terraform", "destroy", "-auto-approve", "-input=false", "-target=proxmox_virtual_environment_vm.vm")
 }
 
+var runTerraform = utils.RunCommandInDirEnv
+
 var webhookTargetTimeout = 70 * time.Minute
 
 var webhookTargetPollInterval = 2 * time.Second
@@ -1283,7 +1282,7 @@ func ensureLifecycleIdentity(terraformDir string, cfg config.Config) error {
 
 func (r Runner) reconcileTerraform(terraformDir string) error {
 	environment := []string{"TF_VAR_proxmox_api_token=" + r.Config.Talos.Proxmox.APITokenID + "=" + r.Config.Talos.Proxmox.APITokenSecret}
-	if err := utils.RunCommandInDirEnv(terraformDir, environment, "terraform", "init", "-input=false"); err != nil {
+	if err := runTerraform(terraformDir, environment, "terraform", "init", "-input=false"); err != nil {
 		return err
 	}
 	if err := importNetworkBridges(terraformDir, environment, r.Config); err != nil {
@@ -1293,19 +1292,41 @@ func (r Runner) reconcileTerraform(terraformDir string) error {
 		if err := r.rebuildTalosVMs(terraformDir, environment); err != nil {
 			return err
 		}
+	} else if err := r.importConfiguredTalosVMs(terraformDir, environment); err != nil {
+		return err
 	}
 	planPath, err := terraformPlanPath(terraformDir, r.Config.ClusterID)
 	if err != nil {
 		return err
 	}
 	defer os.Remove(planPath)
-	if err := utils.RunCommandInDirEnv(terraformDir, environment, "terraform", "plan", "-input=false", "-out="+planPath); err != nil {
+	planArgs := []string{"plan", "-input=false"}
+	if r.Mode == Reconcile {
+		planArgs = append(planArgs, terraformReconcileTargets(r.Config)...)
+	}
+	planArgs = append(planArgs, "-out="+planPath)
+	if err := runTerraform(terraformDir, environment, "terraform", planArgs...); err != nil {
 		return err
 	}
-	return utils.RunCommandInDirEnv(terraformDir, environment, "terraform", "apply", "-input=false", "-auto-approve", planPath)
+	return runTerraform(terraformDir, environment, "terraform", "apply", "-input=false", "-auto-approve", planPath)
+}
+
+func terraformReconcileTargets(cfg config.Config) []string {
+	targets := []string{"-target=proxmox_virtual_environment_download_file.talos_iso"}
+	if cfg.Talos.Cluster.ManageNetworkBridges {
+		targets = append(targets, "-target=proxmox_virtual_environment_network_linux_bridge.cluster_bridge")
+	}
+	return targets
 }
 
 func (r Runner) rebuildTalosVMs(terraformDir string, environment []string) error {
+	if err := r.importConfiguredTalosVMs(terraformDir, environment); err != nil {
+		return err
+	}
+	return destroyTalosVMs(terraformDir, environment)
+}
+
+func (r Runner) importConfiguredTalosVMs(terraformDir string, environment []string) error {
 	if err := verifyTalosVMs(r.Config); err != nil {
 		return err
 	}
@@ -1326,7 +1347,7 @@ func (r Runner) rebuildTalosVMs(terraformDir string, environment []string) error
 	if err := verifyStateTalosVMs(resources, r.Config, true); err != nil {
 		return err
 	}
-	return destroyTalosVMs(terraformDir, environment)
+	return nil
 }
 
 type terraformImport struct {
@@ -1403,7 +1424,7 @@ func importTerraformResources(terraformDir string, environment []string, imports
 		if resources[resource.Address] {
 			continue
 		}
-		if err := utils.RunCommandInDirEnv(terraformDir, environment, "terraform", "import", "-input=false", resource.Address, resource.ID); err != nil {
+		if err := runTerraform(terraformDir, environment, "terraform", "import", "-input=false", resource.Address, resource.ID); err != nil {
 			return fmt.Errorf("import Terraform resource: %w", err)
 		}
 	}

internal/bootstrap/bootstrap_test.go

@@ -911,11 +911,22 @@ func TestTerraformPlanPathIsAbsolute(t *testing.T) {
 	}
 }
 
-func TestTerraformStateResourcesAllowsNoStateFile(t *testing.T) {
+func TestTerraformReconcileTargetsStageImageAndManagedBridgesOnly(t *testing.T) {
+	if targets := strings.Join(terraformReconcileTargets(config.Config{}), " "); targets != "-target=proxmox_virtual_environment_download_file.talos_iso" {
+		t.Fatalf("default reconcile targets = %q", targets)
+	}
+	cfg := config.Config{Talos: config.TalosConfig{Cluster: config.TalosClusterConfig{ManageNetworkBridges: true}}}
+	targets := strings.Join(terraformReconcileTargets(cfg), " ")
+	if !strings.Contains(targets, "-target=proxmox_virtual_environment_download_file.talos_iso") || !strings.Contains(targets, "-target=proxmox_virtual_environment_network_linux_bridge.cluster_bridge") || strings.Contains(targets, "proxmox_virtual_environment_vm.vm") {
+		t.Fatalf("managed bridge reconcile targets = %q", targets)
+	}
+}
+
+func TestTerraformStateResourcesAllowsANSINoStateFile(t *testing.T) {
 	original := runTerraformStateList
 	t.Cleanup(func() { runTerraformStateList = original })
 	runTerraformStateList = func(string, []string) ([]byte, []byte, error) {
-		return nil, []byte("No state file was found!\n"), errors.New("exit status 1")
+		return nil, []byte("\x1b[31mError:\x1b[0m \x1b[31mNO STATE\x1b[0m FILE was found!\n"), errors.New("exit status 1")
 	}
 
 	resources, err := listTerraformStateResources("terraform", nil)
@@ -946,6 +957,130 @@ func TestVerifyStateTalosVMsRejectsForeignAndMissingVMs(t *testing.T) {
 	}
 }
 
+func TestReconcileTerraformImportsConfiguredTalosVMsBeforePlan(t *testing.T) {
+	originalVerify := verifyTalosVMs
+	originalStateList := runTerraformStateList
+	originalRun := runTerraform
+	t.Cleanup(func() {
+		verifyTalosVMs = originalVerify
+		runTerraformStateList = originalStateList
+		runTerraform = originalRun
+	})
+
+	cfg := config.Config{ClusterID: "test-cluster", Talos: config.TalosConfig{
+		Nodes: []config.TalosNode{{Name: "cp-01", ProxmoxNode: "pve", VMID: 100, Role: "controlplane"}},
+	}}
+	importedAddress := talosVMImports(cfg)[0].Address
+	var events, state []string
+	verifyTalosVMs = func(got config.Config) error {
+		if len(got.Talos.Nodes) != 1 || got.Talos.Nodes[0].Name != "cp-01" || got.Talos.Nodes[0].ProxmoxNode != "pve" || got.Talos.Nodes[0].VMID != 100 {
+			t.Fatal("Talos VM identity verification used the wrong config")
+		}
+		events = append(events, "verify")
+		return nil
+	}
+	runTerraformStateList = func(string, []string) ([]byte, []byte, error) {
+		return []byte(strings.Join(state, "\n")), nil, nil
+	}
+	runTerraform = func(_ string, _ []string, name string, args ...string) error {
+		if name != "terraform" {
+			t.Fatalf("unexpected command %q", name)
+		}
+		switch args[0] {
+		case "init":
+			events = append(events, "init")
+		case "import":
+			if len(args) != 4 || args[2] != importedAddress || args[3] != "pve/100" {
+				t.Fatalf("unexpected Talos VM import: %q", args)
+			}
+			events = append(events, "import")
+			state = append(state, importedAddress)
+		case "plan":
+			if len(state) != 1 || state[0] != importedAddress {
+				return errors.New("VM create collision")
+			}
+			if strings.Contains(strings.Join(args, " "), "proxmox_virtual_environment_vm.vm") || !strings.Contains(strings.Join(args, " "), "-target=proxmox_virtual_environment_download_file.talos_iso") {
+				t.Fatalf("reconcile plan did not exclude the Talos VM: %q", args)
+			}
+			events = append(events, "plan")
+		case "apply":
+			if strings.Contains(strings.Join(args, " "), "proxmox_virtual_environment_vm.vm") {
+				t.Fatalf("reconcile apply did not exclude the Talos VM: %q", args)
+			}
+			events = append(events, "apply")
+		default:
+			t.Fatalf("unexpected Terraform operation %q", args[0])
+		}
+		return nil
+	}
+
+	if err := (Runner{Config: cfg, Mode: Reconcile}).reconcileTerraform(t.TempDir()); err != nil {
+		t.Fatal(err)
+	}
+	if strings.Join(events, ",") != "init,verify,import,plan,apply" {
+		t.Fatalf("Terraform phase order = %q", events)
+	}
+}
+
+func TestRebuildTerraformRetainsFullTalosVMLifecycle(t *testing.T) {
+	originalVerify := verifyTalosVMs
+	originalStateList := runTerraformStateList
+	originalRun := runTerraform
+	originalDestroy := destroyTalosVMs
+	t.Cleanup(func() {
+		verifyTalosVMs = originalVerify
+		runTerraformStateList = originalStateList
+		runTerraform = originalRun
+		destroyTalosVMs = originalDestroy
+	})
+
+	cfg := config.Config{ClusterID: "test-cluster", Talos: config.TalosConfig{
+		Nodes: []config.TalosNode{{Name: "cp-01", ProxmoxNode: "pve", VMID: 100, Role: "controlplane"}},
+	}}
+	importedAddress := talosVMImports(cfg)[0].Address
+	var events, state []string
+	verifyTalosVMs = func(config.Config) error {
+		events = append(events, "verify")
+		return nil
+	}
+	runTerraformStateList = func(string, []string) ([]byte, []byte, error) {
+		return []byte(strings.Join(state, "\n")), nil, nil
+	}
+	destroyTalosVMs = func(_ string, _ []string) error {
+		events = append(events, "destroy")
+		return nil
+	}
+	runTerraform = func(_ string, _ []string, name string, args ...string) error {
+		if name != "terraform" {
+			t.Fatalf("unexpected command %q", name)
+		}
+		switch args[0] {
+		case "init":
+			events = append(events, "init")
+		case "import":
+			state = append(state, importedAddress)
+			events = append(events, "import")
+		case "plan":
+			if strings.Contains(strings.Join(args, " "), "-target=") {
+				t.Fatalf("rebuild plan must retain the full VM lifecycle: %q", args)
+			}
+			events = append(events, "plan")
+		case "apply":
+			events = append(events, "apply")
+		default:
+			t.Fatalf("unexpected Terraform operation %q", args[0])
+		}
+		return nil
+	}
+
+	if err := (Runner{Config: cfg, Mode: Rebuild}).reconcileTerraform(t.TempDir()); err != nil {
+		t.Fatal(err)
+	}
+	if strings.Join(events, ",") != "init,verify,import,destroy,plan,apply" {
+		t.Fatalf("Terraform phase order = %q", events)
+	}
+}
+
 func TestEnsureLifecycleIdentityStoresMetadataUnderGeneratedDirectory(t *testing.T) {
 	repo := t.TempDir()
 	terraformDir := filepath.Join(repo, "terraform")