hmdeploy/docker/client.go

package docker

import (
	"context"
	"encoding/json"
	"errors"
	"fmt"
	"os"
	"os/exec"
	"path/filepath"
	"sync"
	"time"

	"gitea.sonak.fr/hmdeploy/connection"
	"gitea.sonak.fr/hmdeploy/models"
	"github.com/rs/zerolog/log"
)

const (
	stateTickDuration   = 4 * time.Second
	defaultStateTimeout = 10 * time.Minute
)

var (
	ErrDockerClientSave = errors.New("unable to save image into tar")

	ErrDockerClientExtractServicesInputLength = errors.New("bad input length")
	ErrDockerClientExtractServicesParse       = errors.New("parse error")

	ErrContextDone = errors.New("unable to execute, context done")
)

type stackOption struct {
	baseDir    string
	checkState bool
}

type fnStackOption func(s *stackOption)

func WithCheckState() fnStackOption {
	return func(s *stackOption) {
		s.checkState = true
	}
}

func WithBaseDir(dir string) fnStackOption {
	return func(s *stackOption) {
		s.baseDir = dir
	}
}

func parseIDs(cmdOutput string) []string {
	ids := []string{}
	bufLine := []rune{}
	for _, c := range cmdOutput {
		if c == '\n' {
			ids = append(ids, string(bufLine))
			bufLine = bufLine[:0]

			continue
		}
		bufLine = append(bufLine, c)
	}
	return ids
}

type IClient interface {
	Save(imageName, dest string) (string, error)
}

// LocalClient is a simple Docker client wrapping the local Docker daemon.
// It does not use the Docker API but instead shell command and collect the output.
//
// NOTE: for now, it's ok, it only needs one command so, no need to add a fat dedicated
// library with full Docker client API.
type LocalClient struct{}

var _ IClient = (*LocalClient)(nil)

func NewLocalClient() LocalClient {
	return LocalClient{}
}

// Save saves the `imageName` (tag included) in tar format in the target directory: `dest`.
// The `dest` directory must exist with correct permissions.
func (c *LocalClient) Save(imageName, dest string) (string, error) {
	destInfo, err := os.Stat(dest)
	if err != nil {
		return "", fmt.Errorf("unable to stat file, dir=%s, err=%v", dest, err)
	}

	if !destInfo.IsDir() {
		return "", fmt.Errorf("dest file must be a directory, dir=%s, err=%v", dest, err)
	}

	tarFile := fmt.Sprintf("%s.tar", imageName)

	cmd := exec.Command("docker", "save", "-o", tarFile, imageName)
	cmd.Dir = dest
	if _, err := cmd.Output(); err != nil {
		return "", fmt.Errorf(
			"%w, dir=%s, image=%s, err=%v",
			ErrDockerClientSave,
			dest,
			imageName,
			err,
		)
	}

	return filepath.Join(dest, tarFile), nil
}

// RemoteClient is a simple Docker client for remote daemon.
// It does not use the Docker API but instead shell command over an SSH connection and collect the output.
//
// NOTE: for now, it's ok, it only needs one command so, no need to add a fat dedicated
// library with full Docker client API.
type RemoteClient struct {
	conn connection.SSHConn
}

func NewRemoteClient(netInfo *models.HMNetInfo) (RemoteClient, error) {
	var rc RemoteClient
	conn, err := connection.NewSSHConn(
		netInfo.IP.String(),
		netInfo.SSH.User,
		netInfo.SSH.Port,
		netInfo.SSH.PrivKey,
	)
	if err != nil {
		return rc, nil
	}

	rc.conn = conn
	return rc, nil
}

type extractOption struct {
	filter string
}

type fnExtractOption func(*extractOption)

func WithName(name string) fnExtractOption {
	return func(o *extractOption) {
		o.filter = name
	}
}

func (c *RemoteClient) getIDS(name string) ([]string, error) {
	cmd := "docker service ls -q"
	if name != "" {
		cmd += " --filter name=" + name
	}

	output, err := c.conn.Execute(cmd)
	if err != nil {
		return nil, err
	}

	return parseIDs(output), nil
}

func (c *RemoteClient) getServiceDetails(id string) (Service, error) {
	output, err := c.conn.Execute(
		fmt.Sprintf(
			`echo "{\"services\": [$(echo $(cmd=$(docker service ps %s --format json); cmd=$(echo $cmd | tr '} {' '},{'); echo $cmd))], \"details\": $(docker service inspect %s --format=json)}"`,
			id,
			id,
		),
	)
	if err != nil {
		return Service{}, err
	}

	sc := Service{}
	if err := json.Unmarshal([]byte(output), &sc); err != nil {
		return sc, err
	}

	return sc, nil
}

func (c *RemoteClient) extractServicesDetails(ids ...string) (Services, error) {
	services := Services{}
	for _, id := range ids {
		srv, err := c.getServiceDetails(id)
		if err != nil {
			return nil, err
		}
		services = append(services, srv)
	}

	return services, nil
}

func (c *RemoteClient) LoadImages(imageNames ...string) error {
	for idx := range imageNames {
		if _, err := c.conn.Execute("docker image load -i " + imageNames[idx] + ".tar"); err != nil {
			return err
		}
	}
	return nil
}

func (c *RemoteClient) DeployStack(
	ctx context.Context,
	projectName, composeFilepath string,
	options ...fnStackOption,
) error {
	var opts stackOption
	for _, opt := range options {
		opt(&opts)
	}

	if bd := opts.baseDir; bd != "" {
		composeFilepath = filepath.Join(bd, composeFilepath)
	}

	if _, err := c.conn.Execute(fmt.Sprintf("docker stack deploy -c %s %s --with-registry-auth", composeFilepath, projectName)); err != nil {
		return err
	}

	if opts.checkState {
		return c.checkState(ctx, projectName, Running)
	}

	return nil
}

func (c *RemoteClient) DestroyStack(
	ctx context.Context,
	projectName string,
	options ...fnStackOption,
) error {
	if _, err := c.conn.Execute(fmt.Sprintf("docker stack rm %s", projectName)); err != nil {
		return err
	}

	var opts stackOption
	for _, opt := range options {
		opt(&opts)
	}

	if opts.checkState {
		return c.checkState(ctx, projectName, Shutdown)
	}

	return nil
}

func (c *RemoteClient) ExtractServicesDetails(options ...fnExtractOption) (Services, error) {
	var opts extractOption
	for _, opt := range options {
		opt(&opts)
	}

	ids, err := c.getIDS(opts.filter)
	if err != nil {
		return nil, err
	}

	return c.extractServicesDetails(ids...)
}

// checkState checks the state of the deployment.
// It loops over all the services deployed for the project (replicas included) and
// checks if the `target` state match the services states.
//
// There's a timeout (default: 30s) that you can set with the options: `WithTimeout`.
func (c *RemoteClient) checkState(
	ctx context.Context,
	projectName string,
	target ServiceStatus,
) error {
	var checkErr error
	var wg sync.WaitGroup
	wg.Add(1)
	go func() {
		defer wg.Done()

		ticker := time.NewTicker(stateTickDuration)
		ctxTimeout, fnCancel := context.WithDeadline(ctx, time.Now().UTC().Add(defaultStateTimeout))
		defer fnCancel()

		for {
			select {
			case <-ticker.C:
				ticker.Stop()
				log.Info().
					Str("project", projectName).
					Str("state", string(target)).
					Msg("checking project state...")
				srvs, err := c.ExtractServicesDetails(WithName(projectName))
				if err != nil {
					checkErr = err
					return
				}

				ready := true
			mainloop:
				for idx := range srvs {
					// ensure app name is the right one
					if srvs[idx].App != projectName {
						continue
					}
					for idy := range srvs[idx].Replicas {
						if srvs[idx].Replicas[idy].State != target {
							log.Info().Dur("retry (ms)", stateTickDuration).Msg("project not in good state yet, retrying...")
							ready = false
							break mainloop
						}
					}
				}
				if ready {
					return
				}

				ticker.Reset(stateTickDuration)
			case <-ctxTimeout.Done():
				msg := "swarm deployment skipped"
				if errors.Is(ctxTimeout.Err(), context.DeadlineExceeded) {
					msg = "swarm check state timeout"
				}
				checkErr = fmt.Errorf("%w, %s", ErrContextDone, msg)
				return
			}
		}
	}()

	wg.Wait()
	return checkErr
}