container: add checkpoint restore support to container start

add checkpoint restore support to container start.
e.g.:
$ nerdctl run --name cr -d busybox sleep infinity
$ nerdctl checkpoint create cr checkpoint1
$ nerdctl start --checkpoint checkpoint  cr

Signed-off-by: ChengyuZhu6 <[email protected]>

Author: ChengyuZhu6

cmd/nerdctl/compose/compose_start.go

@@ -114,7 +114,7 @@ func startContainers(ctx context.Context, client *containerd.Client, containers
 			}
 
 			// in compose, always disable attach
-			if err := containerutil.Start(ctx, c, false, false, client, "", (*config.Config)(globalOptions)); err != nil {
+			if err := containerutil.Start(ctx, c, false, false, client, "", "", (*config.Config)(globalOptions)); err != nil {
 				return err
 			}
 			info, err := c.Info(ctx, containerd.WithoutRefreshedMetadata)

cmd/nerdctl/container/container_run.go

@@ -432,7 +432,7 @@ func runAction(cmd *cobra.Command, args []string) error {
 	logURI := lab[labels.LogURI]
 	detachC := make(chan struct{})
 	task, err := taskutil.NewTask(ctx, client, c, createOpt.Attach, createOpt.Interactive, createOpt.TTY, createOpt.Detach,
-		con, logURI, createOpt.DetachKeys, createOpt.GOptions.Namespace, detachC)
+		con, logURI, createOpt.DetachKeys, createOpt.GOptions.Namespace, detachC, "")
 	if err != nil {
 		return err
 	}

cmd/nerdctl/container/container_start.go

@@ -44,6 +44,8 @@ func StartCommand() *cobra.Command {
 	cmd.Flags().BoolP("attach", "a", false, "Attach STDOUT/STDERR and forward signals")
 	cmd.Flags().String("detach-keys", consoleutil.DefaultDetachKeys, "Override the default detach keys")
 	cmd.Flags().BoolP("interactive", "i", false, "Attach container's STDIN")
+	cmd.Flags().String("checkpoint", "", "checkpoint name")
+	cmd.Flags().String("checkpoint-dir", "", "checkpoint directory")
 	return cmd
 }
 
@@ -64,12 +66,22 @@ func startOptions(cmd *cobra.Command) (types.ContainerStartOptions, error) {
 	if err != nil {
 		return types.ContainerStartOptions{}, err
 	}
+	checkpoint, err := cmd.Flags().GetString("checkpoint")
+	if err != nil {
+		return types.ContainerStartOptions{}, err
+	}
+	checkpointDir, err := cmd.Flags().GetString("checkpoint-dir")
+	if err != nil {
+		return types.ContainerStartOptions{}, err
+	}
 	return types.ContainerStartOptions{
-		Stdout:      cmd.OutOrStdout(),
-		GOptions:    globalOptions,
-		Attach:      attach,
-		DetachKeys:  detachKeys,
-		Interactive: interactive,
+		Stdout:        cmd.OutOrStdout(),
+		GOptions:      globalOptions,
+		Attach:        attach,
+		DetachKeys:    detachKeys,
+		Interactive:   interactive,
+		Checkpoint:    checkpoint,
+		CheckpointDir: checkpointDir,
 	}, nil
 }
 

pkg/api/types/container_types.go

@@ -32,6 +32,10 @@ type ContainerStartOptions struct {
 	DetachKeys string
 	// Attach stdin
 	Interactive bool
+	// Checkpoint is the name of the checkpoint to restore
+	Checkpoint string
+	// CheckpointDir is the directory to store checkpoints
+	CheckpointDir string
 }
 
 // ContainerKillOptions specifies options for `nerdctl (container) kill`.

pkg/cmd/checkpoint/create.go

@@ -21,6 +21,7 @@ import (
 	"encoding/json"
 	"errors"
 	"fmt"
+	"path/filepath"
 
 	ocispec "github.com/opencontainers/image-spec/specs-go/v1"
 
@@ -97,6 +98,9 @@ func Create(ctx context.Context, client *containerd.Client, containerID string,
 		return errors.New("invalid checkpoint")
 	}
 
+	if options.CheckpointDir == "" {
+		options.CheckpointDir = filepath.Join(options.GOptions.DataRoot, "checkpoints")
+	}
 	targetPath, err := checkpointutil.GetCheckpointDir(options.CheckpointDir, checkpointName, container.ID(), true)
 	if err != nil {
 		return err

pkg/cmd/container/restart.go

@@ -48,7 +48,7 @@ func Restart(ctx context.Context, client *containerd.Client, containers []string
 			if err := containerutil.Stop(ctx, found.Container, options.Timeout, options.Signal); err != nil {
 				return err
 			}
-			if err := containerutil.Start(ctx, found.Container, false, false, client, "", (*config.Config)(&options.GOption)); err != nil {
+			if err := containerutil.Start(ctx, found.Container, false, false, client, "", "", (*config.Config)(&options.GOption)); err != nil {
 				return err
 			}
 			_, err = fmt.Fprintln(options.Stdout, found.Req)

pkg/cmd/container/start.go

@@ -19,10 +19,12 @@ package container
 import (
 	"context"
 	"fmt"
+	"path/filepath"
 
 	containerd "github.com/containerd/containerd/v2/client"
 
 	"github.com/containerd/nerdctl/v2/pkg/api/types"
+	"github.com/containerd/nerdctl/v2/pkg/checkpointutil"
 	"github.com/containerd/nerdctl/v2/pkg/config"
 	"github.com/containerd/nerdctl/v2/pkg/containerutil"
 	"github.com/containerd/nerdctl/v2/pkg/idutil/containerwalker"
@@ -33,15 +35,28 @@ func Start(ctx context.Context, client *containerd.Client, reqs []string, option
 	if options.Attach && len(reqs) > 1 {
 		return fmt.Errorf("you cannot start and attach multiple containers at once")
 	}
+	if options.Checkpoint != "" && len(reqs) > 1 {
+		return fmt.Errorf("you cannot start multiple containers with checkpoint at once")
+	}
 
 	walker := &containerwalker.ContainerWalker{
 		Client: client,
 		OnFound: func(ctx context.Context, found containerwalker.Found) error {
 			var err error
+			var checkpointDir string
 			if found.MatchCount > 1 {
 				return fmt.Errorf("multiple IDs found with provided prefix: %s", found.Req)
 			}
-			if err := containerutil.Start(ctx, found.Container, options.Attach, options.Interactive, client, options.DetachKeys, (*config.Config)(&options.GOptions)); err != nil {
+			if options.Checkpoint != "" {
+				if options.CheckpointDir == "" {
+					options.CheckpointDir = filepath.Join(options.GOptions.DataRoot, "checkpoints")
+				}
+				checkpointDir, err = checkpointutil.GetCheckpointDir(options.CheckpointDir, options.Checkpoint, found.Container.ID(), false)
+				if err != nil {
+					return err
+				}
+			}
+			if err := containerutil.Start(ctx, found.Container, options.Attach, options.Interactive, client, options.DetachKeys, checkpointDir, (*config.Config)(&options.GOptions)); err != nil {
 				return err
 			}
 			if !options.Attach {

pkg/containerutil/containerutil.go

@@ -206,7 +206,7 @@ func GenerateSharingPIDOpts(ctx context.Context, targetCon containerd.Container)
 }
 
 // Start starts `container` with `attach` flag. If `attach` is true, it will attach to the container's stdio.
-func Start(ctx context.Context, container containerd.Container, isAttach bool, isInteractive bool, client *containerd.Client, detachKeys string, cfg *config.Config) (err error) {
+func Start(ctx context.Context, container containerd.Container, isAttach bool, isInteractive bool, client *containerd.Client, detachKeys string, checkpointDir string, cfg *config.Config) (err error) {
 	// defer the storage of start error in the dedicated label
 	defer func() {
 		if err != nil {
@@ -280,7 +280,7 @@ func Start(ctx context.Context, container containerd.Container, isAttach bool, i
 		// source: https://github.com/containerd/nerdctl/blob/main/docs/command-reference.md#whale-nerdctl-start
 		attachStreamOpt = []string{"STDOUT", "STDERR"}
 	}
-	task, err := taskutil.NewTask(ctx, client, container, attachStreamOpt, isInteractive, isTerminal, true, con, logURI, detachKeys, namespace, detachC)
+	task, err := taskutil.NewTask(ctx, client, container, attachStreamOpt, isInteractive, isTerminal, true, con, logURI, detachKeys, namespace, detachC, checkpointDir)
 	if err != nil {
 		return err
 	}

pkg/taskutil/taskutil.go

@@ -19,6 +19,7 @@ package taskutil
 import (
 	"context"
 	"errors"
+	"fmt"
 	"io"
 	"net/url"
 	"os"
@@ -27,13 +28,20 @@ import (
 	"strings"
 	"sync"
 	"syscall"
+	"time"
 
 	"github.com/Masterminds/semver/v3"
+	"github.com/opencontainers/go-digest"
 	"golang.org/x/term"
 
 	"github.com/containerd/console"
+	"github.com/containerd/containerd/api/types"
 	containerd "github.com/containerd/containerd/v2/client"
+	"github.com/containerd/containerd/v2/core/content"
+	"github.com/containerd/containerd/v2/core/images"
+	"github.com/containerd/containerd/v2/pkg/archive"
 	"github.com/containerd/containerd/v2/pkg/cio"
+	"github.com/containerd/errdefs"
 	"github.com/containerd/log"
 
 	"github.com/containerd/nerdctl/v2/pkg/cioutil"
@@ -43,9 +51,50 @@ import (
 
 // NewTask is from https://github.com/containerd/containerd/blob/v1.4.3/cmd/ctr/commands/tasks/tasks_unix.go#L70-L108
 func NewTask(ctx context.Context, client *containerd.Client, container containerd.Container,
-	attachStreamOpt []string, isInteractive, isTerminal, isDetach bool, con console.Console, logURI, detachKeys, namespace string, detachC chan<- struct{}) (containerd.Task, error) {
+	attachStreamOpt []string, isInteractive, isTerminal, isDetach bool, con console.Console, logURI, detachKeys, namespace string, detachC chan<- struct{}, checkpointDir string) (containerd.Task, error) {
+	var (
+		checkpoint *types.Descriptor
+		t          containerd.Task
+		err        error
+	)
 
-	var t containerd.Task
+	if checkpointDir != "" {
+		tar := archive.Diff(ctx, "", checkpointDir)
+		cs := client.ContentStore()
+		writer, err := cs.Writer(ctx, content.WithRef(checkpointDir))
+		if err != nil {
+			return nil, err
+		}
+		defer writer.Close()
+		size, err := io.Copy(writer, tar)
+		if err != nil {
+			return nil, err
+		}
+		labels := map[string]string{
+			"containerd.io/gc.root": time.Now().UTC().Format(time.RFC3339),
+		}
+		if err = writer.Commit(ctx, size, "", content.WithLabels(labels)); err != nil {
+			if !errors.Is(err, errdefs.ErrAlreadyExists) {
+				return nil, err
+			}
+		}
+		checkpoint = &types.Descriptor{
+			MediaType: images.MediaTypeContainerd1Checkpoint,
+			Digest:    writer.Digest().String(),
+			Size:      size,
+		}
+		defer func() {
+			if checkpoint != nil {
+				_ = cs.Delete(ctx, digest.Digest(checkpoint.Digest))
+			}
+		}()
+		if err = tar.Close(); err != nil {
+			return nil, fmt.Errorf("failed to close checkpoint tar stream: %w", err)
+		}
+		if err != nil {
+			return nil, fmt.Errorf("failed to upload checkpoint to containerd: %w", err)
+		}
+	}
 	closer := func() {
 		if detachC != nil {
 			detachC <- struct{}{}
@@ -158,7 +207,15 @@ func NewTask(ctx context.Context, client *containerd.Client, container container
 		}
 		ioCreator = cioutil.NewContainerIO(namespace, logURI, false, in, os.Stdout, os.Stderr)
 	}
-	t, err := container.NewTask(ctx, ioCreator)
+
+	taskOpts := []containerd.NewTaskOpts{
+		func(_ context.Context, _ *containerd.Client, info *containerd.TaskInfo) error {
+			info.Checkpoint = checkpoint
+			return nil
+		},
+	}
+
+	t, err = container.NewTask(ctx, ioCreator, taskOpts...)
 	if err != nil {
 		return nil, err
 	}