lightning: revert custom gRPC resolver

br/pkg/lightning/backend/local/local.go

@@ -49,7 +49,6 @@ import (
 	"github.com/pingcap/tidb/br/pkg/logutil"
 	"github.com/pingcap/tidb/br/pkg/membuf"
 	"github.com/pingcap/tidb/br/pkg/pdutil"
-	"github.com/pingcap/tidb/br/pkg/resolver"
 	split "github.com/pingcap/tidb/br/pkg/restore"
 	"github.com/pingcap/tidb/br/pkg/utils"
 	"github.com/pingcap/tidb/br/pkg/version"
@@ -128,22 +127,23 @@ type importClientFactoryImpl struct {
 	conns          *common.GRPCConns
 	splitCli       split.SplitClient
 	tls            *common.TLS
-	resolveBuilder *resolver.Builder
 	tcpConcurrency int
 }
 
 func newImportClientFactoryImpl(splitCli split.SplitClient, tls *common.TLS, tcpConcurrency int) *importClientFactoryImpl {
-	resolveBuilder := resolver.NewBuilder(splitCli)
 	return &importClientFactoryImpl{
 		conns:          common.NewGRPCConns(),
 		splitCli:       splitCli,
 		tls:            tls,
-		resolveBuilder: resolveBuilder,
 		tcpConcurrency: tcpConcurrency,
 	}
 }
 
 func (f *importClientFactoryImpl) makeConn(ctx context.Context, storeID uint64) (*grpc.ClientConn, error) {
+	store, err := f.splitCli.GetStore(ctx, storeID)
+	if err != nil {
+		return nil, errors.Trace(err)
+	}
 	opt := grpc.WithInsecure()
 	if f.tls.TLSConfig() != nil {
 		opt = grpc.WithTransportCredentials(credentials.NewTLS(f.tls.TLSConfig()))
@@ -152,17 +152,21 @@ func (f *importClientFactoryImpl) makeConn(ctx context.Context, storeID uint64)
 
 	bfConf := backoff.DefaultConfig
 	bfConf.MaxDelay = gRPCBackOffMaxDelay
+	// we should use peer address for tiflash. for tikv, peer address is empty
+	addr := store.GetPeerAddress()
+	if addr == "" {
+		addr = store.GetAddress()
+	}
 	conn, err := grpc.DialContext(
 		ctx,
-		f.resolveBuilder.Target(storeID),
+		addr,
 		opt,
 		grpc.WithConnectParams(grpc.ConnectParams{Backoff: bfConf}),
 		grpc.WithKeepaliveParams(keepalive.ClientParameters{
 			Time:                gRPCKeepAliveTime,
 			Timeout:             gRPCKeepAliveTimeout,
 			PermitWithoutStream: true,
 		}),
-		grpc.WithResolvers(f.resolveBuilder),
 	)
 	cancel()
 	if err != nil {

br/pkg/lightning/backend/local/localhelper_test.go

@@ -287,8 +287,6 @@ func (c *testClient) SetStoresLabel(ctx context.Context, stores []uint64, labelK
 	return nil
 }
 
-func (c *testClient) InvalidateStoreCache(storeID uint64) {}
-
 func cloneRegion(region *restore.RegionInfo) *restore.RegionInfo {
 	r := &metapb.Region{}
 	if region.Region != nil {

br/pkg/lightning/restore/restore.go

@@ -367,7 +367,7 @@ func NewRestoreControllerWithPauser(
 
 		backend, err = local.NewLocalBackend(ctx, tls, cfg, g, maxOpenFiles, errorMgr)
 		if err != nil {
-			return nil, errors.Annotate(err, "build local backend failed")
+			return nil, common.NormalizeOrWrapErr(common.ErrUnknown, err)
 		}
 		err = verifyLocalFile(ctx, cpdb, cfg.TikvImporter.SortedKVDir)
 		if err != nil {

br/pkg/restore/split_client.go

@@ -39,7 +39,6 @@ import (
 
 const (
 	splitRegionMaxRetryTime = 4
-	defaultDialTimeout      = time.Minute
 )
 
 // SplitClient is an external client used by RegionSplitter.
@@ -76,8 +75,6 @@ type SplitClient interface {
 	// SetStoresLabel add or update specified label of stores. If labelValue
 	// is empty, it clears the label.
 	SetStoresLabel(ctx context.Context, stores []uint64, labelKey, labelValue string) error
-	// InvalidateStoreCache invalidate store cache for the given store id.
-	InvalidateStoreCache(storeID uint64)
 }
 
 // pdClient is a wrapper of pd client, can be used by RegionSplitter.
@@ -195,7 +192,11 @@ func (c *pdClient) SplitRegion(ctx context.Context, regionInfo *RegionInfo, key
 		peer = regionInfo.Region.Peers[0]
 	}
 	storeID := peer.GetStoreId()
-	conn, err := c.dialStore(ctx, storeID)
+	store, err := c.GetStore(ctx, storeID)
+	if err != nil {
+		return nil, errors.Trace(err)
+	}
+	conn, err := grpc.Dial(store.GetAddress(), grpc.WithInsecure())
 	if err != nil {
 		return nil, errors.Trace(err)
 	}
@@ -311,7 +312,15 @@ func (c *pdClient) sendSplitRegionRequest(
 			peer = regionInfo.Region.Peers[0]
 		}
 		storeID := peer.GetStoreId()
-		conn, err := c.dialStore(ctx, storeID)
+		store, err := c.GetStore(ctx, storeID)
+		if err != nil {
+			return nil, multierr.Append(splitErrors, err)
+		}
+		opt := grpc.WithInsecure()
+		if c.tlsConf != nil {
+			opt = grpc.WithTransportCredentials(credentials.NewTLS(c.tlsConf))
+		}
+		conn, err := grpc.Dial(store.GetAddress(), opt)
 		if err != nil {
 			return nil, multierr.Append(splitErrors, err)
 		}
@@ -368,25 +377,6 @@ func (c *pdClient) sendSplitRegionRequest(
 	return nil, errors.Trace(splitErrors)
 }
 
-func (c *pdClient) dialStore(ctx context.Context, storeID uint64) (*grpc.ClientConn, error) {
-	store, err := c.GetStore(ctx, storeID)
-	if err != nil {
-		return nil, errors.Trace(err)
-	}
-	credsOpt := grpc.WithInsecure()
-	if c.tlsConf != nil {
-		credsOpt = grpc.WithTransportCredentials(credentials.NewTLS(c.tlsConf))
-	}
-	subCtx, cancel := context.WithTimeout(ctx, defaultDialTimeout)
-	defer cancel()
-	conn, err := grpc.DialContext(subCtx, store.GetAddress(), credsOpt, grpc.WithReturnConnectionError())
-	if err != nil {
-		c.InvalidateStoreCache(storeID)
-		return nil, errors.Trace(err)
-	}
-	return conn, nil
-}
-
 func (c *pdClient) BatchSplitRegionsWithOrigin(
 	ctx context.Context, regionInfo *RegionInfo, keys [][]byte,
 ) (*RegionInfo, []*RegionInfo, error) {
@@ -610,12 +600,6 @@ func (c *pdClient) getPDAPIAddr() string {
 	return strings.TrimRight(addr, "/")
 }
 
-func (c *pdClient) InvalidateStoreCache(storeID uint64) {
-	c.mu.Lock()
-	delete(c.storeCache, storeID)
-	c.mu.Unlock()
-}
-
 func checkRegionEpoch(_new, _old *RegionInfo) bool {
 	return _new.Region.GetId() == _old.Region.GetId() &&
 		_new.Region.GetRegionEpoch().GetVersion() == _old.Region.GetRegionEpoch().GetVersion() &&

br/pkg/restore/split_test.go

@@ -237,8 +237,6 @@ func (c *TestClient) SetStoresLabel(ctx context.Context, stores []uint64, labelK
 	return nil
 }
 
-func (c *TestClient) InvalidateStoreCache(storeID uint64) {}
-
 type assertRetryLessThanBackoffer struct {
 	max     int
 	already int

br/tests/br_split_region_fail/run.sh

@@ -52,7 +52,7 @@ BR_LOG_TO_TERM=1
 
 grep "a error occurs on split region" $LOG && \
 grep "split region meet not leader error" $LOG && \
-grep "Full restore success" $LOG && \
+grep "Full Restore success" $LOG && \
 grep "find new leader" $LOG
 
 if [ $? -ne 0 ]; then

br/tests/lightning_fail_fast/run.sh

@@ -16,7 +16,7 @@
 
 set -eux
 
-export GO_FAILPOINTS='github.com/pingcap/tidb/br/pkg/lightning/restore/SlowDownWriteRows=sleep(50);github.com/pingcap/tidb/br/pkg/lightning/restore/SetMinDeliverBytes=return(1)'
+export GO_FAILPOINTS='github.com/pingcap/tidb/br/pkg/lightning/restore/SlowDownWriteRows=sleep(100);github.com/pingcap/tidb/br/pkg/lightning/restore/SetMinDeliverBytes=return(1)'
 
 for CFG in chunk engine; do
   rm -f "$TEST_DIR/lightning-tidb.log"