Lazy Restart

Dockerfile

@@ -4,8 +4,9 @@
 FROM golang:1.26-alpine AS builder
 
 WORKDIR /app
-COPY . .
+COPY go.mod go.sum ./
 RUN go mod download
+COPY . .
 WORKDIR /app/cmd/backup
 RUN go build -o backup .
 

cmd/backup/config.go

@@ -97,6 +97,7 @@ type Config struct {
 	GoogleDriveImpersonateSubject        string          `split_words:"true"`
 	GoogleDriveEndpoint                  string          `split_words:"true"`
 	GoogleDriveTokenURL                  string          `split_words:"true"`
+	ActivateLazyRestart                  bool            `split_words:"true" default:"false"`
 	source                               string
 	additionalEnvVars                    map[string]string
 }

cmd/backup/lock.go

@@ -11,11 +11,13 @@
 	"github.com/offen/docker-volume-backup/internal/errwrap"
 )
 
+const LOCK_FILE = "/var/lock/dockervolumebackup.lock"
+
 // lock opens a lockfile at the given location, keeping it locked until the
 // caller invokes the returned release func. In case the lock is currently blocked
 // by another execution, it will repeatedly retry until the lock is available
 // or the given timeout is exceeded.
-func (s *script) lock(lockfile string) (func() error, error) {
+func (s *script) lock() (func() error, error) {
 	start := time.Now()
 	defer func() {
 		s.stats.LockedTime = time.Since(start)
@@ -26,7 +28,8 @@
 	deadline := time.NewTimer(s.c.LockTimeout)
 	defer deadline.Stop()
 
-	fileLock := flock.New(lockfile)
+	fileLock := flock.New(LOCK_FILE)
+	defer s.removeWaitingFile()
 
 	for {
 		acquired, err := fileLock.TryLock()
@@ -41,6 +44,7 @@
 		}
 
 		if !s.encounteredLock {
+			s.writeWaitingFile()
 			s.logger.Info(
 				fmt.Sprintf(
 					"Exclusive lock was not available on first attempt. Will retry until it becomes available or the timeout of %s is exceeded.",

cmd/backup/run_script.go

@@ -32,17 +32,6 @@ func runScript(c *Config) (err error) {
 
 	s := newScript(c)
 
-	unlock, lockErr := s.lock("/var/lock/dockervolumebackup.lock")
-	if lockErr != nil {
-		err = errwrap.Wrap(lockErr, "error acquiring file lock")
-		return
-	}
-	defer func() {
-		if derr := unlock(); derr != nil {
-			err = errors.Join(err, errwrap.Wrap(derr, "error releasing file lock"))
-		}
-	}()
-
 	unset, warnings, err := s.c.resolve()
 	if err != nil {
 		return errwrap.Wrap(err, "error applying env")
@@ -81,6 +70,17 @@ func runScript(c *Config) (err error) {
 		return
 	}
 
+	unlock, lockErr := s.lock()
+	if lockErr != nil {
+		err = errwrap.Wrap(lockErr, "error acquiring file lock")
+		return
+	}
+	defer func() {
+		if derr := unlock(); derr != nil {
+			err = errors.Join(err, errwrap.Wrap(derr, "error releasing file lock"))
+		}
+	}()
+
 	err = func() (err error) {
 		scriptErr := func() error {
 			if err := s.withLabeledCommands(lifecyclePhaseArchive, func() (err error) {

cmd/backup/script.go

@@ -11,6 +11,7 @@ import (
 	"text/template"
 	"time"
 
+	"github.com/google/uuid"
 	"github.com/offen/docker-volume-backup/internal/errwrap"
 	"github.com/offen/docker-volume-backup/internal/storage"
 	"github.com/offen/docker-volume-backup/internal/storage/azure"
@@ -30,6 +31,8 @@ import (
 // script holds all the stateful information required to orchestrate a
 // single backup run.
 type script struct {
+	id string
+
 	cli       *client.Client
 	storages  []storage.Backend
 	logger    *slog.Logger
@@ -43,6 +46,10 @@ type script struct {
 
 	encounteredLock bool
 
+	isSwarm             bool
+	containersToStop    []handledContainer
+	servicesToScaleDown []handledSwarmService
+
 	c *Config
 }
 
@@ -72,6 +79,8 @@ func newScript(c *Config) *script {
 }
 
 func (s *script) init() error {
+	s.id = uuid.New().String()
+
 	s.registerHook(hookLevelPlumbing, func(error) error {
 		s.stats.EndTime = time.Now()
 		s.stats.TookTime = s.stats.EndTime.Sub(s.stats.StartTime)
@@ -135,6 +144,14 @@ func (s *script) init() error {
 			}
 			return nil
 		})
+		s.isSwarm, err = isSwarm(s.cli)
+		if err != nil {
+			return errwrap.Wrap(err, "error determining swarm state")
+		}
+		err = s.determineContainersAndServicesToStop()
+		if err != nil {
+			return errwrap.Wrap(err, "error determining containers and services to stop")
+		}
 	}
 
 	logFunc := func(logType storage.LogLevel, context string, msg string, params ...any) {