feat: add directory watching and auto-upload functionality

client/middleware/default.go

@@ -15,7 +15,7 @@ import (
 // https://github.com/iyear/tdl/blob/master/core/tclient/tclient.go
 func NewDefaultMiddlewares(ctx context.Context, timeout time.Duration) []telegram.Middleware {
 	return []telegram.Middleware{
-		recovery.New(ctx, newBackoff(timeout)),
+		recovery.New(ctx, func() backoff.BackOff { return newBackoff(timeout) }),
 		retry.New(config.C().Telegram.RpcRetry),
 		floodwait.NewSimpleWaiter(),
 	}

client/middleware/recovery/recovery.go

@@ -14,19 +14,28 @@ import (
 )
 
 type recovery struct {
-	ctx     context.Context
-	backoff backoff.BackOff
+	ctx        context.Context
+	newBackoff func() backoff.BackOff
 }
 
-func New(ctx context.Context, backoff backoff.BackOff) telegram.Middleware {
+// New returns a recovery middleware.
+//
+// newBackoff is a factory that must return a fresh backoff.BackOff on every call: backoff implementations in
+// cenkalti/backoff/v4 (notably ExponentialBackOff) are not safe for concurrent
+// use, and the Telegram client invokes RPCs from many goroutines in parallel.
+//
+// Sharing a single instance corrupts its internal counters, breaks the
+// exponential interval, and defeats MaxElapsedTime - see issue #218.
+func New(ctx context.Context, newBackoff func() backoff.BackOff) telegram.Middleware {
 	return &recovery{
-		ctx:     ctx,
-		backoff: backoff,
+		ctx:        ctx,
+		newBackoff: newBackoff,
 	}
 }
 
 func (r *recovery) Handle(next tg.Invoker) telegram.InvokeFunc {
 	return func(ctx context.Context, input bin.Encoder, output bin.Decoder) error {
+		b := r.newBackoff()
 
 		return backoff.RetryNotify(func() error {
 			if err := next.Invoke(ctx, input, output); err != nil {
@@ -38,7 +47,7 @@ func (r *recovery) Handle(next tg.Invoker) telegram.InvokeFunc {
 			}
 
 			return nil
-		}, r.backoff, func(err error, duration time.Duration) {
+		}, b, func(err error, duration time.Duration) {
 			log.FromContext(ctx).Debug("Wait for connection recovery", "error", err, "duration", duration)
 		})
 	}

cmd/root.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 
 	"github.com/krau/SaveAny-Bot/cmd/upload"
+	"github.com/krau/SaveAny-Bot/cmd/watch"
 	"github.com/krau/SaveAny-Bot/config"
 	"github.com/spf13/cobra"
 )
@@ -18,6 +19,7 @@ var rootCmd = &cobra.Command{
 func init() {
 	config.RegisterFlags(rootCmd)
 	upload.Register(rootCmd)
+	watch.Register(rootCmd)
 }
 
 func Execute(ctx context.Context) {

cmd/watch/cmd.go

@@ -0,0 +1,145 @@
+package watch
+
+import (
+	"fmt"
+	"time"
+
+	"github.com/charmbracelet/log"
+	"github.com/krau/SaveAny-Bot/client/bot"
+	"github.com/krau/SaveAny-Bot/common/cache"
+	"github.com/krau/SaveAny-Bot/common/i18n"
+	"github.com/krau/SaveAny-Bot/common/utils/tgutil"
+	"github.com/krau/SaveAny-Bot/config"
+	"github.com/krau/SaveAny-Bot/database"
+	stortype "github.com/krau/SaveAny-Bot/pkg/enums/storage"
+	"github.com/krau/SaveAny-Bot/storage"
+	"github.com/spf13/cobra"
+)
+
+var watchCmd = &cobra.Command{
+	Use:   "watch",
+	Short: "watch a local directory and auto-upload changed files to storage",
+	Long: `Watch a local directory and automatically upload created or modified files
+to the specified storage backend, preserving the relative directory structure.
+
+Example:
+  saveany-bot watch -p /data/inbox -s mystorage -d backup --recursive`,
+	RunE: runWatch,
+}
+
+func Register(root *cobra.Command) {
+	flags := watchCmd.Flags()
+	flags.StringP("path", "p", "", "local directory path to watch")
+	watchCmd.MarkFlagRequired("path")
+	flags.StringP("storage", "s", "", "storage name to upload to")
+	watchCmd.MarkFlagRequired("storage")
+	flags.StringP("dir", "d", "", "storage dir to upload to, default is the base_path of the storage")
+	flags.BoolP("recursive", "r", false, "watch subdirectories recursively")
+	flags.Bool("overwrite", false, "overwrite existing files on storage instead of skipping")
+	flags.Bool("initial-scan", false, "upload existing files in the directory on startup")
+	flags.Duration("debounce", 2*time.Second, "wait time after the last change before uploading a file")
+	flags.Int("upload-workers", 0, "number of concurrent uploads, default is config.workers")
+	flags.Duration("retry-delay", 3*time.Second, "delay between upload retries")
+	root.AddCommand(watchCmd)
+}
+
+func runWatch(cmd *cobra.Command, _ []string) error {
+	watchPath, err := cmd.Flags().GetString("path")
+	if err != nil {
+		return err
+	}
+	storName, err := cmd.Flags().GetString("storage")
+	if err != nil {
+		return err
+	}
+	destDir, err := cmd.Flags().GetString("dir")
+	if err != nil {
+		return err
+	}
+	recursive, err := cmd.Flags().GetBool("recursive")
+	if err != nil {
+		return err
+	}
+	overwrite, err := cmd.Flags().GetBool("overwrite")
+	if err != nil {
+		return err
+	}
+	initialScan, err := cmd.Flags().GetBool("initial-scan")
+	if err != nil {
+		return err
+	}
+	debounce, err := cmd.Flags().GetDuration("debounce")
+	if err != nil {
+		return err
+	}
+	uploadWorkers, err := cmd.Flags().GetInt("upload-workers")
+	if err != nil {
+		return err
+	}
+	retryDelay, err := cmd.Flags().GetDuration("retry-delay")
+	if err != nil {
+		return err
+	}
+
+	ctx := cmd.Context()
+	logger := log.FromContext(ctx)
+
+	configFile := config.GetConfigFile(cmd)
+	if err := config.Init(ctx, configFile); err != nil {
+		return fmt.Errorf("failed to load config: %w", err)
+	}
+	i18n.Init(config.C().Lang)
+	cache.Init()
+	database.Init(ctx)
+
+	stor, err := storage.GetStorageByName(ctx, storName)
+	if err != nil {
+		return fmt.Errorf("failed to get storage %q: %w", storName, err)
+	}
+
+	// Telegram storage needs the bot client and its ext context injected into ctx.
+	if stor.Type() == stortype.Telegram {
+		bot.Init(ctx)
+		ctx = tgutil.ExtWithContext(ctx, bot.ExtContext())
+	}
+
+	if uploadWorkers < 1 {
+		uploadWorkers = config.C().Workers
+	}
+
+	uploader := NewUploader(ctx, UploaderOptions{
+		Storage:    stor,
+		DestDir:    destDir,
+		Overwrite:  overwrite,
+		Workers:    uploadWorkers,
+		Retry:      config.C().Retry,
+		RetryDelay: retryDelay,
+	})
+
+	watcher, err := NewWatcher(ctx, WatcherOptions{
+		Root:      watchPath,
+		Recursive: recursive,
+		Debounce:  debounce,
+		Uploader:  uploader,
+	})
+	if err != nil {
+		uploader.Close()
+		return fmt.Errorf("failed to create watcher: %w", err)
+	}
+
+	if initialScan {
+		watcher.ScanExisting(ctx)
+	}
+
+	logger.Infof("watch started: %s -> storage %q dir %q", watchPath, storName, destDir)
+
+	// Run blocks until ctx is cancelled (e.g. SIGINT).
+	runErr := watcher.Run(ctx)
+
+	// Wait for in-flight uploads to finish before exiting.
+	logger.Info("waiting for in-flight uploads to finish...")
+	uploader.Close()
+	logger.Info("watch stopped")
+
+	return runErr
+}

cmd/watch/uploader.go

@@ -0,0 +1,227 @@
+package watch
+
+import (
+	"context"
+	"fmt"
+	"os"
+	"path"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"github.com/charmbracelet/log"
+	"github.com/krau/SaveAny-Bot/pkg/enums/ctxkey"
+	"github.com/krau/SaveAny-Bot/storage"
+)
+
+type uploadJob struct {
+	// localPath is the absolute path of the local file.
+	localPath string
+	// relPath is relative to the watch root, used to preserve directory structure on storage.
+	relPath string
+}
+
+// Uploader uploads local files to the target storage via a worker pool.
+// If a file changes while being uploaded, it is re-uploaded once after the
+// current upload finishes, instead of being queued multiple times.
+type Uploader struct {
+	stor       storage.Storage
+	destDir    string
+	overwrite  bool
+	retry      int
+	retryDelay time.Duration
+	logger     *log.Logger
+
+	jobs chan uploadJob
+	wg   sync.WaitGroup
+
+	mu sync.Mutex
+	// inflight maps in-progress (or queued) file paths. A true value means the
+	// file changed during upload and must be re-queued once done.
+	inflight map[string]bool
+}
+
+type UploaderOptions struct {
+	Storage    storage.Storage
+	DestDir    string
+	Overwrite  bool
+	Workers    int
+	Retry      int
+	RetryDelay time.Duration
+	QueueSize  int
+}
+
+// NewUploader creates and starts an Uploader. The caller must call Close when done.
+func NewUploader(ctx context.Context, opts UploaderOptions) *Uploader {
+	if opts.Workers < 1 {
+		opts.Workers = 1
+	}
+	if opts.Retry < 1 {
+		opts.Retry = 1
+	}
+	if opts.RetryDelay <= 0 {
+		opts.RetryDelay = 3 * time.Second
+	}
+	if opts.QueueSize < opts.Workers {
+		opts.QueueSize = opts.Workers * 64
+	}
+
+	u := &Uploader{
+		stor:       opts.Storage,
+		destDir:    opts.DestDir,
+		overwrite:  opts.Overwrite,
+		retry:      opts.Retry,
+		retryDelay: opts.RetryDelay,
+		logger:     log.FromContext(ctx).WithPrefix("uploader"),
+		jobs:       make(chan uploadJob, opts.QueueSize),
+		inflight:   make(map[string]bool),
+	}
+
+	for i := 0; i < opts.Workers; i++ {
+		u.wg.Add(1)
+		go u.worker(ctx)
+	}
+
+	return u
+}
+
+// Submit enqueues an upload job. If the same file is already in flight, it is
+// marked for re-upload instead of being queued again. Returns false if ctx is
+// cancelled before the job can be enqueued.
+func (u *Uploader) Submit(ctx context.Context, job uploadJob) bool {
+	u.mu.Lock()
+	if _, ok := u.inflight[job.localPath]; ok {
+		u.inflight[job.localPath] = true
+		u.mu.Unlock()
+		u.logger.Debugf("file %s already in flight, marked for re-upload", job.localPath)
+		return true
+	}
+	u.inflight[job.localPath] = false
+	u.mu.Unlock()
+
+	select {
+	case u.jobs <- job:
+		return true
+	case <-ctx.Done():
+		u.mu.Lock()
+		delete(u.inflight, job.localPath)
+		u.mu.Unlock()
+		return false
+	}
+}
+
+func (u *Uploader) worker(ctx context.Context) {
+	defer u.wg.Done()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case job, ok := <-u.jobs:
+			if !ok {
+				return
+			}
+			u.process(ctx, job)
+		}
+	}
+}
+
+func (u *Uploader) process(ctx context.Context, job uploadJob) {
+	if err := u.uploadWithRetry(ctx, job); err != nil {
+		if ctx.Err() != nil {
+			u.clearInflight(job.localPath)
+			return
+		}
+		u.logger.Errorf("failed to upload %s after %d attempt(s): %v", job.localPath, u.retry, err)
+	}
+
+	// Re-queue if the file changed again while it was being uploaded.
+	u.mu.Lock()
+	needReupload := u.inflight[job.localPath]
+	if needReupload {
+		u.inflight[job.localPath] = false
+	} else {
+		delete(u.inflight, job.localPath)
+	}
+	u.mu.Unlock()
+
+	if needReupload {
+		select {
+		case u.jobs <- job:
+			u.logger.Debugf("re-queued %s due to changes during upload", job.localPath)
+		case <-ctx.Done():
+			u.clearInflight(job.localPath)
+		}
+	}
+}
+
+func (u *Uploader) clearInflight(localPath string) {
+	u.mu.Lock()
+	delete(u.inflight, localPath)
+	u.mu.Unlock()
+}
+
+func (u *Uploader) uploadWithRetry(ctx context.Context, job uploadJob) error {
+	var lastErr error
+	for attempt := 1; attempt <= u.retry; attempt++ {
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
+		err := u.upload(ctx, job)
+		if err == nil {
+			return nil
+		}
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
+		lastErr = err
+		u.logger.Warnf("upload %s failed (attempt %d/%d): %v", job.localPath, attempt, u.retry, err)
+		if attempt < u.retry {
+			select {
+			case <-time.After(u.retryDelay):
+			case <-ctx.Done():
+				return ctx.Err()
+			}
+		}
+	}
+	return lastErr
+}
+
+func (u *Uploader) upload(ctx context.Context, job uploadJob) error {
+	file, err := os.Open(filepath.Clean(job.localPath))
+	if err != nil {
+		return fmt.Errorf("failed to open file: %w", err)
+	}
+	defer file.Close()
+
+	info, err := file.Stat()
+	if err != nil {
+		return fmt.Errorf("failed to stat file: %w", err)
+	}
+	if info.IsDir() {
+		return fmt.Errorf("path is a directory, not a file")
+	}
+
+	// Keep the relative directory structure on the storage side.
+	storagePath := path.Join(u.destDir, filepath.ToSlash(job.relPath))
+
+	uploadCtx := context.WithValue(ctx, ctxkey.ContentLength, info.Size())
+	if u.overwrite {
+		uploadCtx = storage.WithOverwrite(uploadCtx)
+	} else if u.stor.Exists(uploadCtx, storagePath) {
+		u.logger.Infof("skip existing file: %s", storagePath)
+		return nil
+	}
+
+	u.logger.Infof("uploading %s -> %s (%d bytes)", job.localPath, storagePath, info.Size())
+	if err := u.stor.Save(uploadCtx, file, storagePath); err != nil {
+		return fmt.Errorf("failed to save to storage: %w", err)
+	}
+	u.logger.Infof("uploaded %s", storagePath)
+	return nil
+}
+
+// Close stops accepting jobs and waits for in-flight uploads to finish.
+func (u *Uploader) Close() {
+	close(u.jobs)
+	u.wg.Wait()
+}

cmd/watch/watcher.go

@@ -0,0 +1,269 @@
+package watch
+
+import (
+	"context"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"sync"
+	"time"
+
+	"github.com/charmbracelet/log"
+	"github.com/fsnotify/fsnotify"
+)
+
+// Watcher watches a local directory and submits stable files to the Uploader.
+//
+// Write-completion detection: fsnotify emits Write events throughout a write.
+// Watcher debounces per file and only uploads once the file size stays
+// unchanged across a debounce window, avoiding uploads of partial files.
+type Watcher struct {
+	root      string
+	recursive bool
+	debounce  time.Duration
+	uploader  *Uploader
+	logger    *log.Logger
+
+	fsw *fsnotify.Watcher
+
+	mu      sync.Mutex
+	pending map[string]*time.Timer
+	// lastSize is the last observed file size, used to detect a stable write.
+	lastSize map[string]int64
+}
+
+type WatcherOptions struct {
+	Root      string
+	Recursive bool
+	Debounce  time.Duration
+	Uploader  *Uploader
+}
+
+// NewWatcher creates a Watcher.
+func NewWatcher(ctx context.Context, opts WatcherOptions) (*Watcher, error) {
+	if opts.Debounce <= 0 {
+		opts.Debounce = 2 * time.Second
+	}
+	root, err := filepath.Abs(opts.Root)
+	if err != nil {
+		return nil, fmt.Errorf("failed to resolve root path: %w", err)
+	}
+	info, err := os.Stat(root)
+	if err != nil {
+		return nil, fmt.Errorf("failed to stat root path: %w", err)
+	}
+	if !info.IsDir() {
+		return nil, fmt.Errorf("watch path must be a directory: %s", root)
+	}
+
+	fsw, err := fsnotify.NewWatcher()
+	if err != nil {
+		return nil, fmt.Errorf("failed to create fsnotify watcher: %w", err)
+	}
+
+	w := &Watcher{
+		root:      root,
+		recursive: opts.Recursive,
+		debounce:  opts.Debounce,
+		uploader:  opts.Uploader,
+		logger:    log.FromContext(ctx).WithPrefix("watcher"),
+		fsw:       fsw,
+		pending:   make(map[string]*time.Timer),
+		lastSize:  make(map[string]int64),
+	}
+	return w, nil
+}
+
+// Run starts watching and blocks until ctx is cancelled.
+func (w *Watcher) Run(ctx context.Context) error {
+	if err := w.addDir(w.root); err != nil {
+		w.fsw.Close()
+		return fmt.Errorf("failed to watch root: %w", err)
+	}
+	w.logger.Infof("watching %s (recursive=%v, debounce=%s)", w.root, w.recursive, w.debounce)
+
+	defer w.cleanup()
+
+	for {
+		select {
+		case <-ctx.Done():
+			w.logger.Info("stopping watcher")
+			return nil
+		case event, ok := <-w.fsw.Events:
+			if !ok {
+				return nil
+			}
+			w.handleEvent(ctx, event)
+		case err, ok := <-w.fsw.Errors:
+			if !ok {
+				return nil
+			}
+			w.logger.Errorf("watch error: %v", err)
+		}
+	}
+}
+
+func (w *Watcher) handleEvent(ctx context.Context, event fsnotify.Event) {
+	// Remove/Rename: cancel any pending upload for this path.
+	if event.Has(fsnotify.Remove) || event.Has(fsnotify.Rename) {
+		w.cancelPending(event.Name)
+		return
+	}
+
+	if !event.Has(fsnotify.Create) && !event.Has(fsnotify.Write) {
+		return
+	}
+
+	info, err := os.Stat(event.Name)
+	if err != nil {
+		// File may have been removed or moved; ignore.
+		return
+	}
+
+	if info.IsDir() {
+		// New directory: watch it recursively and scan files already inside.
+		if event.Has(fsnotify.Create) && w.recursive {
+			if err := w.addDir(event.Name); err != nil {
+				w.logger.Errorf("failed to watch new dir %s: %v", event.Name, err)
+			}
+			w.scanExisting(ctx, event.Name)
+		}
+		return
+	}
+
+	w.scheduleUpload(ctx, event.Name)
+}
+
+// scheduleUpload schedules a debounced upload for a file.
+func (w *Watcher) scheduleUpload(ctx context.Context, file string) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+
+	if t, ok := w.pending[file]; ok {
+		t.Stop()
+	}
+	w.pending[file] = time.AfterFunc(w.debounce, func() {
+		w.maybeUpload(ctx, file)
+	})
+}
+
+// maybeUpload submits the upload once the debounce window passes and the file
+// size is stable; otherwise it waits another window.
+func (w *Watcher) maybeUpload(ctx context.Context, file string) {
+	if ctx.Err() != nil {
+		return
+	}
+
+	info, err := os.Stat(file)
+	if err != nil {
+		w.cancelPending(file)
+		return
+	}
+	if info.IsDir() {
+		w.cancelPending(file)
+		return
+	}
+
+	w.mu.Lock()
+	prevSize, seen := w.lastSize[file]
+	curSize := info.Size()
+	if !seen || prevSize != curSize {
+		// Size still changing: likely still being written, wait another window.
+		w.lastSize[file] = curSize
+		w.pending[file] = time.AfterFunc(w.debounce, func() {
+			w.maybeUpload(ctx, file)
+		})
+		w.mu.Unlock()
+		return
+	}
+	// Size stable: treat write as complete.
+	delete(w.pending, file)
+	delete(w.lastSize, file)
+	w.mu.Unlock()
+
+	relPath, err := filepath.Rel(w.root, file)
+	if err != nil {
+		w.logger.Errorf("failed to compute relative path for %s: %v", file, err)
+		return
+	}
+
+	w.uploader.Submit(ctx, uploadJob{localPath: file, relPath: relPath})
+}
+
+func (w *Watcher) cancelPending(file string) {
+	w.mu.Lock()
+	defer w.mu.Unlock()
+	if t, ok := w.pending[file]; ok {
+		t.Stop()
+		delete(w.pending, file)
+	}
+	delete(w.lastSize, file)
+}
+
+// addDir adds a directory to the watch list, recursively when enabled.
+func (w *Watcher) addDir(dir string) error {
+	if !w.recursive {
+		return w.fsw.Add(dir)
+	}
+	return filepath.WalkDir(dir, func(p string, d fs.DirEntry, err error) error {
+		if err != nil {
+			w.logger.Warnf("skip path %s: %v", p, err)
+			return nil
+		}
+		if d.IsDir() {
+			if addErr := w.fsw.Add(p); addErr != nil {
+				w.logger.Warnf("failed to watch dir %s: %v", p, addErr)
+			} else {
+				w.logger.Debugf("watching dir %s", p)
+			}
+		}
+		return nil
+	})
+}
+
+// scanExisting submits files already present under dir (initial sync and new-dir backfill).
+func (w *Watcher) scanExisting(ctx context.Context, dir string) {
+	walkFn := func(p string, d fs.DirEntry, err error) error {
+		if err != nil {
+			w.logger.Warnf("skip path %s: %v", p, err)
+			return nil
+		}
+		if ctx.Err() != nil {
+			return ctx.Err()
+		}
+		if d.IsDir() {
+			if !w.recursive && p != dir {
+				return fs.SkipDir
+			}
+			return nil
+		}
+		relPath, relErr := filepath.Rel(w.root, p)
+		if relErr != nil {
+			w.logger.Errorf("failed to compute relative path for %s: %v", p, relErr)
+			return nil
+		}
+		w.uploader.Submit(ctx, uploadJob{localPath: p, relPath: relPath})
+		return nil
+	}
+	if err := filepath.WalkDir(dir, walkFn); err != nil && ctx.Err() == nil {
+		w.logger.Errorf("failed to scan dir %s: %v", dir, err)
+	}
+}
+
+// ScanExisting triggers a one-time scan and upload of existing files under the watch root.
+func (w *Watcher) ScanExisting(ctx context.Context) {
+	w.logger.Info("scanning existing files for initial sync")
+	w.scanExisting(ctx, w.root)
+}
+
+func (w *Watcher) cleanup() {
+	w.mu.Lock()
+	for _, t := range w.pending {
+		t.Stop()
+	}
+	w.pending = make(map[string]*time.Timer)
+	w.lastSize = make(map[string]int64)
+	w.mu.Unlock()
+	w.fsw.Close()
+}

go.mod

@@ -117,7 +117,7 @@ require (
 	github.com/dgraph-io/ristretto/v2 v2.4.0
 	github.com/dop251/goja v0.0.0-20260311135729-065cd970411c
 	github.com/duke-git/lancet/v2 v2.3.9
-	github.com/fsnotify/fsnotify v1.9.0 // indirect
+	github.com/fsnotify/fsnotify v1.9.0
 	github.com/glebarez/sqlite v1.11.0
 	github.com/inconshreveable/mousetrap v1.1.0 // indirect
 	github.com/klauspost/compress v1.18.5 // indirect