br: stop log when full restore failed

br/pkg/restore/client.go

@@ -1520,12 +1520,14 @@ LOOPFORTABLE:
 			restoreFn := func() error {
 				filesGroups := getGroupFiles(filesReplica, rc.fileImporter.supportMultiIngest)
 				for _, filesGroup := range filesGroups {
-					if importErr := func(fs []*backuppb.File) error {
+					if importErr := func(fs []*backuppb.File) (err error) {
 						fileStart := time.Now()
 						defer func() {
-							log.Info("import files done", logutil.Files(filesGroup),
-								zap.Duration("take", time.Since(fileStart)))
-							updateCh.Inc()
+							if err == nil {
+								log.Info("import files done", logutil.Files(filesGroup),
+									zap.Duration("take", time.Since(fileStart)))
+								updateCh.Inc()
+							}
 						}()
 						return rc.fileImporter.ImportSSTFiles(ectx, fs, rewriteRules, rc.cipher, rc.dom.Store().GetCodec().GetAPIVersion())
 					}(filesGroup); importErr != nil {

br/pkg/restore/import.go

@@ -1186,13 +1186,11 @@ func (importer *FileImporter) downloadRawKVSSTV2(
 }
 
 func (importer *FileImporter) ingest(
-	c context.Context,
+	ctx context.Context,
 	files []*backuppb.File,
 	info *split.RegionInfo,
 	downloadMetas []*import_sstpb.SSTMeta,
 ) error {
-	ctx, cancel := context.WithTimeout(c, gRPCTimeOut)
-	defer cancel()
 	for {
 		ingestResp, errIngest := importer.ingestSSTs(ctx, downloadMetas, info)
 		if errIngest != nil {

br/pkg/utils/backoff.go

@@ -31,9 +31,9 @@ const (
 	backupSSTWaitInterval    = 2 * time.Second
 	backupSSTMaxWaitInterval = 3 * time.Second
 
-	resetTSRetryTime       = 16
+	resetTSRetryTime       = 32
 	resetTSWaitInterval    = 50 * time.Millisecond
-	resetTSMaxWaitInterval = 500 * time.Millisecond
+	resetTSMaxWaitInterval = 2 * time.Second
 
 	resetTSRetryTimeExt       = 600
 	resetTSWaitIntervalExt    = 500 * time.Millisecond
@@ -167,7 +167,6 @@ func NewBackupSSTBackoffer() Backoffer {
 }
 
 func (bo *importerBackoffer) NextBackoff(err error) time.Duration {
-	log.Warn("retry to import ssts", zap.Int("attempt", bo.attempt), zap.Error(err))
 	// we don't care storeID here.
 	res := bo.errContext.HandleErrorMsg(err.Error(), 0)
 	if res.Strategy == RetryStrategy {
@@ -249,8 +248,12 @@ func (bo *pdReqBackoffer) NextBackoff(err error) time.Duration {
 		bo.delayTime = 2 * bo.delayTime
 		bo.attempt--
 	default:
+		// If the connection timeout, pd client would cancel the context, and return grpc context cancel error.
+		// So make the codes.Canceled retryable too.
+		// It's OK to retry the grpc context cancel error, because the parent context cancel returns context.Canceled.
+		// For example, cancel the `ectx` and then pdClient.GetTS(ectx) returns context.Canceled instead of grpc context canceled.
 		switch status.Code(e) {
-		case codes.DeadlineExceeded, codes.NotFound, codes.AlreadyExists, codes.PermissionDenied, codes.ResourceExhausted, codes.Aborted, codes.OutOfRange, codes.Unavailable, codes.DataLoss, codes.Unknown:
+		case codes.DeadlineExceeded, codes.Canceled, codes.NotFound, codes.AlreadyExists, codes.PermissionDenied, codes.ResourceExhausted, codes.Aborted, codes.OutOfRange, codes.Unavailable, codes.DataLoss, codes.Unknown:
 			bo.delayTime = 2 * bo.delayTime
 			bo.attempt--
 		default:

br/pkg/utils/backoff_test.go

@@ -123,26 +123,20 @@ func TestPdBackoffWithRetryableError(t *testing.T) {
 		if counter == 2 {
 			return io.EOF
 		}
+		if counter == 6 {
+			return context.Canceled
+		}
 		return gRPCError
 	}, backoffer)
-	require.Equal(t, 16, counter)
+	require.Equal(t, 7, counter)
 	require.Equal(t, []error{
 		gRPCError,
 		gRPCError,
 		io.EOF,
 		gRPCError,
 		gRPCError,
 		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
-		gRPCError,
+		context.Canceled,
 	}, multierr.Errors(err))
 }