Remove idle connection timeout handling

Remove error handling for closed connections

Add progress watchdog to detect stalled accounts

Improve error modal styling and pagination

internal/orchestrator/orchestrator.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"log/slog"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"github.com/vasyansk/imap-copier/internal/crypto"
@@ -22,6 +23,16 @@ var ErrAlreadyRunning = errors.New("task already running")
 // suppressed" note.
 const maxAccountErrors = 500
 
+// A running account that emits no scan/copy progress for stallTimeout is wedged
+// (silent server mid-FETCH, stalled APPEND). The watchdog cancels it so the
+// connections close and the worker unwinds instead of hanging forever. This
+// replaces socket-level read deadlines, which can't tell an idle connection
+// from a stuck one. The threshold is generous so slow-but-live runs aren't cut.
+const (
+	stallTimeout       = 3 * time.Minute
+	stallCheckInterval = 30 * time.Second
+)
+
 // folderPlan is one source folder scheduled for copy and its destination name.
 type folderPlan struct {
 	src, dst string
@@ -316,7 +327,34 @@ func (o *Orchestrator) runAccount(ctx context.Context, task store.Task, runID in
 		_ = dst.Close()
 	}()
 
+	// Progress watchdog: track the last time we saw scan/copy activity; if it
+	// goes quiet for stallTimeout, cancel the account so the connections close
+	// and this worker unwinds (it would otherwise block forever on a silent
+	// server). touch() is called on every progress signal below.
+	var lastActivity atomic.Int64
+	lastActivity.Store(time.Now().UnixNano())
+	touch := func() { lastActivity.Store(time.Now().UnixNano()) }
+	go func() {
+		t := time.NewTicker(stallCheckInterval)
+		defer t.Stop()
+		for {
+			select {
+			case <-actx.Done():
+				return
+			case <-t.C:
+				if time.Since(time.Unix(0, lastActivity.Load())) > stallTimeout {
+					slog.Warn("account stalled with no progress; cancelling",
+						"account", a.ID, "src_login", a.SrcLogin, "stall", stallTimeout)
+					_ = o.store.SetAccountError(ctx, a.ID, "stalled: no progress for "+stallTimeout.String()+", cancelled")
+					cancel()
+					return
+				}
+			}
+		}
+	}()
+
 	folders, err := imapx.ListFolders(src)
+	touch()
 	if err != nil {
 		return o.accountFailed(ctx, task.ID, runID, a, srcEP, dstEP, "src", err)
 	}
@@ -335,6 +373,7 @@ func (o *Orchestrator) runAccount(ctx context.Context, task store.Task, runID in
 		}
 		plan[i].total = n
 		grandTotal += n
+		touch()
 	}
 	o.hub.Publish(wshub.Event{Type: "plan", TaskID: task.ID, Data: map[string]any{
 		"account_id": a.ID, "src_login": a.SrcLogin, "folders": len(plan), "total": grandTotal,
@@ -367,6 +406,7 @@ func (o *Orchestrator) runAccount(ctx context.Context, task store.Task, runID in
 		MarkMigrated: func(folder, k string) error { return o.store.MarkMigrated(ctx, a.ID, folder, k) },
 		OnError:      func(ref, msg string) { addErr("message", curFolder, ref, msg) },
 		OnProgress: func(c, s int) {
+			touch()
 			now := time.Now()
 			done := c + s
 			// throttle to ~3/sec per account, but always emit folder completion
@@ -386,6 +426,7 @@ func (o *Orchestrator) runAccount(ctx context.Context, task store.Task, runID in
 		},
 		// Fires after EXAMINE (before the long fetch) with the folder's message count.
 		OnFolder: func(srcFolder, dstFolder string, total int64) {
+			touch()
 			curFolder, curTotal = srcFolder, total
 			o.hub.Publish(wshub.Event{Type: "folder", TaskID: task.ID, Data: map[string]any{
 				"account_id": a.ID, "src_login": a.SrcLogin,
@@ -395,6 +436,7 @@ func (o *Orchestrator) runAccount(ctx context.Context, task store.Task, runID in
 		// Fires while streaming metadata (dedup scan) so the UI shows movement
 		// before bodies start copying. Throttled to ~4/sec, always emit the last.
 		OnScan: func(scanned, total int64) {
+			touch()
 			now := time.Now()
 			if now.Sub(lastScanEmit) < 250*time.Millisecond && scanned < total {
 				return