feat: add command to restart GCP VMs and implement related functionality

NOTICE

@@ -1163,15 +1163,15 @@ License URL: https://github.com/gomodules/jsonpatch/blob/v2.5.0/v2/LICENSE
 
 ----------
 Module: google.golang.org/api
-Version: v0.273.0
+Version: v0.273.1
 License: BSD-3-Clause
-License URL: https://github.com/googleapis/google-api-go-client/blob/v0.273.0/LICENSE
+License URL: https://github.com/googleapis/google-api-go-client/blob/v0.273.1/LICENSE
 
 ----------
 Module: google.golang.org/api/internal/third_party/uritemplates
-Version: v0.273.0
+Version: v0.273.1
 License: BSD-3-Clause
-License URL: https://github.com/googleapis/google-api-go-client/blob/v0.273.0/internal/third_party/uritemplates/LICENSE
+License URL: https://github.com/googleapis/google-api-go-client/blob/v0.273.1/internal/third_party/uritemplates/LICENSE
 
 ----------
 Module: google.golang.org/genproto/googleapis
@@ -1193,9 +1193,9 @@ License URL: https://github.com/googleapis/go-genproto/blob/d00831a3d3e7/googlea
 
 ----------
 Module: google.golang.org/grpc
-Version: v1.79.3
+Version: v1.80.0
 License: Apache-2.0
-License URL: https://github.com/grpc/grpc-go/blob/v1.79.3/LICENSE
+License URL: https://github.com/grpc/grpc-go/blob/v1.80.0/LICENSE
 
 ----------
 Module: google.golang.org/protobuf

cli/cmd/bootstrap_gcp.go

@@ -105,6 +105,7 @@ func AddBootstrapGcpCmd(parent *cobra.Command, opts *GlobalOptions) {
 	parent.AddCommand(bootstrapGcpCmd.cmd)
 	AddBootstrapGcpPostconfigCmd(bootstrapGcpCmd.cmd, opts)
 	AddBootstrapGcpCleanupCmd(bootstrapGcpCmd.cmd, opts)
+	AddBootstrapGcpRestartVMsCmd(bootstrapGcpCmd.cmd, opts)
 }
 
 func (c *BootstrapGcpCmd) BootstrapGcp() error {

cli/cmd/bootstrap_gcp_restart_vms.go

@@ -0,0 +1,134 @@
+// Copyright (c) Codesphere Inc.
+// SPDX-License-Identifier: Apache-2.0
+
+package cmd
+
+import (
+	"fmt"
+	"log"
+	"os"
+
+	csio "github.com/codesphere-cloud/cs-go/pkg/io"
+	"github.com/codesphere-cloud/oms/internal/bootstrap"
+	"github.com/codesphere-cloud/oms/internal/bootstrap/gcp"
+	"github.com/codesphere-cloud/oms/internal/util"
+	"github.com/spf13/cobra"
+)
+
+type BootstrapGcpRestartVMsCmd struct {
+	cmd  *cobra.Command
+	Opts *BootstrapGcpRestartVMsOpts
+}
+
+type BootstrapGcpRestartVMsOpts struct {
+	*GlobalOptions
+	ProjectID string
+	Zone      string
+	Name      string
+}
+
+func (c *BootstrapGcpRestartVMsCmd) RunE(_ *cobra.Command, args []string) error {
+	ctx := c.cmd.Context()
+	stlog := bootstrap.NewStepLogger(false)
+	fw := util.NewFilesystemWriter()
+
+	projectID := c.Opts.ProjectID
+	zone := c.Opts.Zone
+
+	if projectID == "" || zone == "" {
+		infraFilePath := gcp.GetInfraFilePath()
+		infraEnv, exists, err := gcp.LoadInfraFile(fw, infraFilePath)
+		if err != nil {
+			return fmt.Errorf("failed to load infra file: %w", err)
+		}
+		if !exists {
+			return fmt.Errorf("infra file not found at %s; use --project-id and --zone flags", infraFilePath)
+		}
+		if projectID == "" {
+			projectID = infraEnv.ProjectID
+		}
+		if zone == "" {
+			zone = infraEnv.Zone
+		}
+	}
+
+	if projectID == "" {
+		return fmt.Errorf("project ID is required; set --project-id or ensure the infra file exists")
+	}
+	if zone == "" {
+		return fmt.Errorf("zone is required; set --zone or ensure the infra file exists")
+	}
+
+	gcpClient := gcp.NewGCPClient(ctx, stlog, os.Getenv("GOOGLE_APPLICATION_CREDENTIALS"))
+
+	csEnv := &gcp.CodesphereEnvironment{
+		ProjectID: projectID,
+		Zone:      zone,
+	}
+
+	bs, err := gcp.NewGCPBootstrapper(
+		ctx,
+		nil,
+		stlog,
+		csEnv,
+		nil,
+		gcpClient,
+		fw,
+		nil,
+		nil,
+		util.NewTime(),
+		nil,
+	)
+	if err != nil {
+		return fmt.Errorf("failed to create bootstrapper: %w", err)
+	}
+
+	if c.Opts.Name != "" {
+		log.Printf("Restarting VM %s in project %s (zone %s)...", c.Opts.Name, projectID, zone)
+		if err := bs.RestartVM(c.Opts.Name); err != nil {
+			return fmt.Errorf("failed to restart VM: %w", err)
+		}
+		log.Printf("VM %s restarted successfully.", c.Opts.Name)
+	} else {
+		log.Printf("Restarting all VMs in project %s (zone %s)...", projectID, zone)
+		if err := bs.RestartVMs(); err != nil {
+			return fmt.Errorf("failed to restart VMs: %w", err)
+		}
+		log.Printf("All VMs restarted successfully.")
+	}
+
+	return nil
+}
+
+func AddBootstrapGcpRestartVMsCmd(bootstrapGcp *cobra.Command, opts *GlobalOptions) {
+	restartVMs := BootstrapGcpRestartVMsCmd{
+		cmd: &cobra.Command{
+			Use:   "restart-vms",
+			Short: "Restart stopped or terminated GCP VMs",
+			Long: csio.Long(`Restarts GCP compute instances that were stopped or terminated,
+				for example after spot VM preemption.
+				By default, restarts all VMs defined in the infrastructure.
+				Use --name to restart a single VM.
+				Project ID and zone are read from the local infra file if available,
+				or can be specified via flags.`),
+			Example: formatExamples("beta bootstrap-gcp restart-vms", []csio.Example{
+				{Desc: "Restart all VMs using project info from the local infra file"},
+				{Cmd: "--name jumpbox", Desc: "Restart only the jumpbox VM"},
+				{Cmd: "--name k0s-1", Desc: "Restart a specific k0s node"},
+				{Cmd: "--project-id my-project --zone us-central1-a", Desc: "Restart all VMs with explicit project and zone"},
+				{Cmd: "--project-id my-project --zone us-central1-a --name ceph-1", Desc: "Restart a specific VM with explicit project and zone"},
+			}),
+		},
+		Opts: &BootstrapGcpRestartVMsOpts{
+			GlobalOptions: opts,
+		},
+	}
+
+	flags := restartVMs.cmd.Flags()
+	flags.StringVar(&restartVMs.Opts.ProjectID, "project-id", "", "GCP Project ID (optional, will use infra file if not provided)")
+	flags.StringVar(&restartVMs.Opts.Zone, "zone", "", "GCP Zone (optional, will use infra file if not provided)")
+	flags.StringVar(&restartVMs.Opts.Name, "name", "", "Name of a specific VM to restart (e.g. jumpbox, postgres, ceph-1, k0s-1). Restarts all VMs if not specified.")
+
+	restartVMs.cmd.RunE = restartVMs.RunE
+	bootstrapGcp.AddCommand(restartVMs.cmd)
+}

docs/oms_beta_bootstrap-gcp.md

@@ -63,4 +63,5 @@ oms beta bootstrap-gcp [flags]
 * [oms beta](oms_beta.md)	 - Commands for early testing
 * [oms beta bootstrap-gcp cleanup](oms_beta_bootstrap-gcp_cleanup.md)	 - Clean up GCP infrastructure created by bootstrap-gcp
 * [oms beta bootstrap-gcp postconfig](oms_beta_bootstrap-gcp_postconfig.md)	 - Run post-configuration steps for GCP bootstrapping
+* [oms beta bootstrap-gcp restart-vms](oms_beta_bootstrap-gcp_restart-vms.md)	 - Restart stopped or terminated GCP VMs
 

docs/oms_beta_bootstrap-gcp_restart-vms.md

@@ -0,0 +1,50 @@
+## oms beta bootstrap-gcp restart-vms
+
+Restart stopped or terminated GCP VMs
+
+### Synopsis
+
+Restarts GCP compute instances that were stopped or terminated,
+for example after spot VM preemption.
+By default, restarts all VMs defined in the infrastructure.
+Use --name to restart a single VM.
+Project ID and zone are read from the local infra file if available,
+or can be specified via flags.
+
+```
+oms beta bootstrap-gcp restart-vms [flags]
+```
+
+### Examples
+
+```
+# Restart all VMs using project info from the local infra file
+$ oms beta bootstrap-gcp restart-vms
+
+# Restart only the jumpbox VM
+$ oms beta bootstrap-gcp restart-vms --name jumpbox
+
+# Restart a specific k0s node
+$ oms beta bootstrap-gcp restart-vms --name k0s-1
+
+# Restart all VMs with explicit project and zone
+$ oms beta bootstrap-gcp restart-vms --project-id my-project --zone us-central1-a
+
+# Restart a specific VM with explicit project and zone
+$ oms beta bootstrap-gcp restart-vms --project-id my-project --zone us-central1-a --name ceph-1
+
+```
+
+### Options
+
+```
+  -h, --help                help for restart-vms
+      --name string         Name of a specific VM to restart (e.g. jumpbox, postgres, ceph-1, k0s-1). Restarts all VMs if not specified.
+      --project-id string   GCP Project ID (optional, will use infra file if not provided)
+      --zone string         GCP Zone (optional, will use infra file if not provided)
+```
+
+### SEE ALSO
+
+* [oms beta bootstrap-gcp](oms_beta_bootstrap-gcp.md)	 - Bootstrap GCP infrastructure for Codesphere
+

internal/bootstrap/gcp/gce.go

@@ -366,6 +366,92 @@ func (b *GCPBootstrapper) waitForInstanceRunning(projectID, zone, name string, n
 		name, pollInterval*time.Duration(maxAttempts))
 }
 
+// findVMDef looks up a VM definition by name. Returns the VMDef and true if found.
+func findVMDef(name string) (VMDef, bool) {
+	for _, vm := range vmDefs {
+		if vm.Name == name {
+			return vm, true
+		}
+	}
+	return VMDef{}, false
+}
+
+// validVMNames returns the list of known VM names from vmDefs.
+func validVMNames() []string {
+	names := make([]string, len(vmDefs))
+	for i, vm := range vmDefs {
+		names[i] = vm.Name
+	}
+	return names
+}
+
+// RestartVM restarts a single stopped or terminated VM by a name that is defined in vmDefs.
+func (b *GCPBootstrapper) RestartVM(name string) error {
+	vm, found := findVMDef(name)
+	if !found {
+		return fmt.Errorf("unknown VM name %q; valid names are: %s", name, strings.Join(validVMNames(), ", "))
+	}
+
+	projectID := b.Env.ProjectID
+	zone := b.Env.Zone
+
+	inst, err := b.GCPClient.GetInstance(projectID, zone, name)
+	if err != nil {
+		return fmt.Errorf("failed to get instance %s: %w", name, err)
+	}
+
+	switch s := inst.GetStatus(); s {
+	case "RUNNING":
+		b.stlog.Logf("Instance %s is already running", name)
+		return nil
+	case "TERMINATED", "STOPPED":
+		b.stlog.Logf("Starting stopped instance %s...", name)
+		if err := b.GCPClient.StartInstance(projectID, zone, name); err != nil {
+			return fmt.Errorf("failed to start instance %s: %w", name, err)
+		}
+	case "SUSPENDED":
+		return fmt.Errorf("instance %s is SUSPENDED; manual resume is required", name)
+	default:
+		return fmt.Errorf("instance %s is in unexpected state %q", name, s)
+	}
+
+	readyInstance, err := b.waitForInstanceRunning(projectID, zone, name, vm.ExternalIP)
+	if err != nil {
+		return fmt.Errorf("instance %s did not become ready: %w", name, err)
+	}
+
+	internalIP, externalIP := ExtractInstanceIPs(readyInstance)
+	b.stlog.Logf("Instance %s is now running (internal=%s, external=%s)", name, internalIP, externalIP)
+	return nil
+}
+
+// RestartVMs restarts all stopped or terminated VMs defined in vmDefs.
+func (b *GCPBootstrapper) RestartVMs() error {
+	wg := sync.WaitGroup{}
+	errCh := make(chan error, len(vmDefs))
+
+	for _, vm := range vmDefs {
+		wg.Add(1)
+		go func(vm VMDef) {
+			defer wg.Done()
+			if err := b.RestartVM(vm.Name); err != nil {
+				errCh <- err
+			}
+		}(vm)
+	}
+	wg.Wait()
+	close(errCh)
+
+	var errs []error
+	for err := range errCh {
+		errs = append(errs, err)
+	}
+	if len(errs) > 0 {
+		return fmt.Errorf("errors restarting VMs: %w", errors.Join(errs...))
+	}
+	return nil
+}
+
 // ReadSSHKey reads an SSH key file, expanding ~ in the path
 func (b *GCPBootstrapper) ReadSSHKey(path string) (string, error) {
 	realPath := util.ExpandPath(path)