diff --git a/config/retry/config.go b/config/retry/config.go
index 89a9bc382d..c95d2cbd15 100644
--- a/config/retry/config.go
+++ b/config/retry/config.go
@@ -46,6 +46,7 @@ import (
 	"github.com/tikv/client-go/v2/internal/logutil"
 	"github.com/tikv/client-go/v2/kv"
 	"github.com/tikv/client-go/v2/metrics"
+	"github.com/tikv/client-go/v2/util"
 	"go.uber.org/zap"
 )
 
@@ -191,6 +192,11 @@ func newBackoffFn(base, cap, jitter int) backoffFn {
 		if maxSleepMs >= 0 && realSleep > maxSleepMs {
 			realSleep = maxSleepMs
 		}
+		if _, err := util.EvalFailpoint("fastBackoffBySkipSleep"); err == nil {
+			attempts++
+			lastSleep = sleep
+			return realSleep
+		}
 		select {
 		case <-time.After(time.Duration(realSleep) * time.Millisecond):
 			attempts++
diff --git a/internal/locate/region_cache.go b/internal/locate/region_cache.go
index e8f652e00f..d1b511e256 100644
--- a/internal/locate/region_cache.go
+++ b/internal/locate/region_cache.go
@@ -2843,6 +2843,9 @@ func (s *Store) requestLivenessAndStartHealthCheckLoopIfNeeded(bo *retry.Backoff
 				reResolveInterval = dur
 			}
 		}
+		if _, err := util.EvalFailpoint("skipStoreCheckUntilHealth"); err == nil {
+			return
+		}
 		go s.checkUntilHealth(c, liveness, reResolveInterval)
 	}
 	return
diff --git a/internal/locate/region_request.go b/internal/locate/region_request.go
index efcafbf162..655b47ab97 100644
--- a/internal/locate/region_request.go
+++ b/internal/locate/region_request.go
@@ -73,6 +73,9 @@ import (
 // network error because tidb-server expect tikv client to exit as soon as possible.
 var shuttingDown uint32
 
+// randIntn is only use for testing.
+var randIntn = rand.Intn
+
 // StoreShuttingDown atomically stores ShuttingDown into v.
 func StoreShuttingDown(v uint32) {
 	atomic.StoreUint32(&shuttingDown, v)
@@ -612,7 +615,7 @@ func (state *tryNewProxy) next(bo *retry.Backoffer, selector *replicaSelector) (
 	}
 
 	// Skip advanceCnt valid candidates to find a proxy peer randomly
-	advanceCnt := rand.Intn(candidateNum)
+	advanceCnt := randIntn(candidateNum)
 	for idx, replica := range selector.replicas {
 		if !state.isCandidate(AccessIndex(idx), replica) {
 			continue
@@ -668,13 +671,13 @@ func (state *accessFollower) next(bo *retry.Backoffer, selector *replicaSelector
 	resetStaleRead := false
 	if state.lastIdx < 0 {
 		if state.tryLeader {
-			state.lastIdx = AccessIndex(rand.Intn(replicaSize))
+			state.lastIdx = AccessIndex(randIntn(replicaSize))
 		} else {
 			if replicaSize <= 1 {
 				state.lastIdx = state.leaderIdx
 			} else {
 				// Randomly select a non-leader peer
-				state.lastIdx = AccessIndex(rand.Intn(replicaSize - 1))
+				state.lastIdx = AccessIndex(randIntn(replicaSize - 1))
 				if state.lastIdx >= state.leaderIdx {
 					state.lastIdx++
 				}
@@ -696,7 +699,7 @@ func (state *accessFollower) next(bo *retry.Backoffer, selector *replicaSelector
 	}
 	var offset int
 	if state.lastIdx >= 0 {
-		offset = rand.Intn(replicaSize)
+		offset = randIntn(replicaSize)
 	}
 	reloadRegion := false
 	for i := 0; i < replicaSize && !state.option.leaderOnly; i++ {
@@ -791,16 +794,7 @@ func (state *accessFollower) next(bo *retry.Backoffer, selector *replicaSelector
 }
 
 func (state *accessFollower) IsLeaderExhausted(leader *replica) bool {
-	// Allow another extra retry for the following case:
-	// 1. The stale read is enabled and leader peer is selected as the target peer at first.
-	// 2. Data is not ready is returned from the leader peer.
-	// 3. Stale read flag is removed and processing falls back to snapshot read on the leader peer.
-	// 4. The leader peer should be retried again using snapshot read.
-	if state.isStaleRead && state.option.leaderOnly {
-		return leader.isExhausted(2, 0)
-	} else {
-		return leader.isExhausted(1, 0)
-	}
+	return leader.isExhausted(1, 0)
 }
 
 func (state *accessFollower) onSendFailure(bo *retry.Backoffer, selector *replicaSelector, cause error) {
@@ -845,7 +839,7 @@ func (state *tryIdleReplica) next(bo *retry.Backoffer, selector *replicaSelector
 	// Select a follower replica that has the lowest estimated wait duration
 	minWait := time.Duration(math.MaxInt64)
 	targetIdx := state.leaderIdx
-	startIdx := rand.Intn(len(selector.replicas))
+	startIdx := randIntn(len(selector.replicas))
 	for i := 0; i < len(selector.replicas); i++ {
 		idx := (i + startIdx) % len(selector.replicas)
 		r := selector.replicas[idx]
diff --git a/internal/locate/region_request3_test.go b/internal/locate/region_request3_test.go
index b2ede8ab64..36e1f71bba 100644
--- a/internal/locate/region_request3_test.go
+++ b/internal/locate/region_request3_test.go
@@ -759,6 +759,10 @@ func (s *testRegionRequestToThreeStoresSuite) TestReplicaSelector() {
 // TODO(youjiali1995): Remove duplicated tests. This test may be duplicated with other
 // tests but it's a dedicated one to test sending requests with the replica selector.
 func (s *testRegionRequestToThreeStoresSuite) TestSendReqWithReplicaSelector() {
+	s.NoError(failpoint.Enable("tikvclient/fastBackoffBySkipSleep", `return`))
+	defer func() {
+		s.NoError(failpoint.Disable("tikvclient/fastBackoffBySkipSleep"))
+	}()
 	req := tikvrpc.NewRequest(tikvrpc.CmdRawPut, &kvrpcpb.RawPutRequest{
 		Key:   []byte("key"),
 		Value: []byte("value"),
@@ -987,49 +991,6 @@ func (s *testRegionRequestToThreeStoresSuite) TestSendReqWithReplicaSelector() {
 	for _, store := range s.storeIDs {
 		s.cluster.StartStore(store)
 	}
-
-	// Verify switch to the leader immediately when stale read requests with global txn scope meet region errors.
-	s.cluster.ChangeLeader(region.Region.id, s.peerIDs[0])
-	reachable.injectConstantLiveness(s.cache)
-	s.Eventually(func() bool {
-		stores := s.regionRequestSender.replicaSelector.regionStore.stores
-		return stores[0].getLivenessState() == reachable &&
-			stores[1].getLivenessState() == reachable &&
-			stores[2].getLivenessState() == reachable
-	}, 3*time.Second, 200*time.Millisecond)
-	reloadRegion()
-	req = tikvrpc.NewRequest(tikvrpc.CmdGet, &kvrpcpb.GetRequest{Key: []byte("key")})
-	req.ReadReplicaScope = oracle.GlobalTxnScope
-	req.TxnScope = oracle.GlobalTxnScope
-	for i := 0; i < 10; i++ {
-		req.EnableStaleWithMixedReplicaRead()
-		// The request may be sent to the leader directly. We have to distinguish it.
-		failureOnFollower := 0
-		failureOnLeader := 0
-		s.regionRequestSender.client = &fnClient{fn: func(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (response *tikvrpc.Response, err error) {
-			if addr != s.cluster.GetStore(s.storeIDs[0]).Address {
-				failureOnFollower++
-				return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{RegionError: &errorpb.Error{}}}, nil
-			} else if failureOnLeader == 0 && i%2 == 0 {
-				failureOnLeader++
-				return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{RegionError: &errorpb.Error{}}}, nil
-			} else {
-				return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{}}, nil
-			}
-		}}
-		sender.SendReq(bo, req, region.Region, time.Second)
-		state, ok := sender.replicaSelector.state.(*accessFollower)
-		s.True(ok)
-		s.True(failureOnFollower <= 1) // any retry should go to the leader, hence at most one failure on the follower allowed
-		if failureOnFollower == 0 && failureOnLeader == 0 {
-			// if the request goes to the leader and succeeds then it is executed as a StaleRead
-			s.True(req.StaleRead)
-		} else {
-			// otherwise #leaderOnly flag should be set and retry request as a normal read
-			s.True(state.option.leaderOnly)
-			s.False(req.StaleRead)
-		}
-	}
 }
 
 func (s *testRegionRequestToThreeStoresSuite) TestLoadBasedReplicaRead() {
@@ -1223,62 +1184,6 @@ func (s *testRegionRequestToThreeStoresSuite) TestAccessFollowerAfter1TiKVDown()
 	}
 }
 
-func (s *testRegionRequestToThreeStoresSuite) TestStaleReadFallback() {
-	leaderStore, _ := s.loadAndGetLeaderStore()
-	leaderLabel := []*metapb.StoreLabel{
-		{
-			Key:   "id",
-			Value: strconv.FormatUint(leaderStore.StoreID(), 10),
-		},
-	}
-	regionLoc, err := s.cache.LocateRegionByID(s.bo, s.regionID)
-	s.Nil(err)
-	s.NotNil(regionLoc)
-	value := []byte("value")
-	isFirstReq := true
-
-	s.regionRequestSender.client = &fnClient{fn: func(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (response *tikvrpc.Response, err error) {
-		select {
-		case <-ctx.Done():
-			return nil, errors.New("timeout")
-		default:
-		}
-		// Return `DataIsNotReady` for the first time on leader.
-		if isFirstReq {
-			isFirstReq = false
-			return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{RegionError: &errorpb.Error{
-				DataIsNotReady: &errorpb.DataIsNotReady{},
-			}}}, nil
-		}
-		return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{Value: value}}, nil
-	}}
-
-	region, _ := s.cache.searchCachedRegionByID(regionLoc.Region.GetID())
-	s.True(region.isValid())
-
-	req := tikvrpc.NewReplicaReadRequest(tikvrpc.CmdGet, &kvrpcpb.GetRequest{Key: []byte("key")}, kv.ReplicaReadLeader, nil)
-	req.ReadReplicaScope = oracle.GlobalTxnScope
-	req.TxnScope = oracle.GlobalTxnScope
-	req.EnableStaleWithMixedReplicaRead()
-	req.ReplicaReadType = kv.ReplicaReadMixed
-	var ops []StoreSelectorOption
-	ops = append(ops, WithMatchLabels(leaderLabel))
-
-	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
-	defer cancel()
-	bo := retry.NewBackoffer(ctx, -1)
-	s.Nil(err)
-	resp, _, _, err := s.regionRequestSender.SendReqCtx(bo, req, regionLoc.Region, time.Second, tikvrpc.TiKV, ops...)
-	s.Nil(err)
-
-	regionErr, err := resp.GetRegionError()
-	s.Nil(err)
-	s.Nil(regionErr)
-	getResp, ok := resp.Resp.(*kvrpcpb.GetResponse)
-	s.True(ok)
-	s.Equal(getResp.Value, value)
-}
-
 func (s *testRegionRequestToThreeStoresSuite) TestSendReqFirstTimeout() {
 	leaderAddr := ""
 	reqTargetAddrs := make(map[string]struct{})
diff --git a/internal/locate/region_request_state_test.go b/internal/locate/region_request_state_test.go
index 6a92b52767..c94552b2f2 100644
--- a/internal/locate/region_request_state_test.go
+++ b/internal/locate/region_request_state_test.go
@@ -348,17 +348,6 @@ func TestRegionCacheStaleRead(t *testing.T) {
 			followerSuccessReplica:  []string{"z2"},
 			followerSuccessReadType: SuccessStaleRead,
 		},
-		{
-			do:                      leaderDataIsNotReady,
-			leaderRegionValid:       true,
-			leaderAsyncReload:       Some(false),
-			leaderSuccessReplica:    []string{"z1"},
-			leaderSuccessReadType:   SuccessLeaderRead,
-			followerRegionValid:     true,
-			followerAsyncReload:     Some(false),
-			followerSuccessReplica:  []string{"z2"},
-			followerSuccessReadType: SuccessStaleRead,
-		},
 		{
 			do:                      followerDataIsNotReady,
 			leaderRegionValid:       true,
@@ -395,45 +384,6 @@ func TestRegionCacheStaleRead(t *testing.T) {
 			followerSuccessReplica:  []string{"z1"},
 			followerSuccessReadType: SuccessLeaderRead,
 		},
-		{
-			do:                      leaderDataIsNotReady,
-			extra:                   []func(suite *testRegionCacheStaleReadSuite){followerServerIsBusy},
-			recoverable:             true,
-			leaderRegionValid:       true,
-			leaderAsyncReload:       Some(false),
-			leaderSuccessReplica:    []string{"z1"},
-			leaderSuccessReadType:   SuccessLeaderRead,
-			followerRegionValid:     true,
-			followerAsyncReload:     Some(false),
-			followerSuccessReplica:  []string{"z1"},
-			followerSuccessReadType: SuccessLeaderRead,
-		},
-		{
-			do:                      leaderDataIsNotReady,
-			extra:                   []func(suite *testRegionCacheStaleReadSuite){followerDataIsNotReady},
-			recoverable:             true,
-			leaderRegionValid:       true,
-			leaderAsyncReload:       Some(false),
-			leaderSuccessReplica:    []string{"z1"},
-			leaderSuccessReadType:   SuccessLeaderRead,
-			followerRegionValid:     true,
-			followerAsyncReload:     Some(false),
-			followerSuccessReplica:  []string{"z1"},
-			followerSuccessReadType: SuccessLeaderRead,
-		},
-		{
-			do:                      leaderDataIsNotReady,
-			extra:                   []func(suite *testRegionCacheStaleReadSuite){followerDown},
-			recoverable:             true,
-			leaderRegionValid:       true,
-			leaderAsyncReload:       Some(false),
-			leaderSuccessReplica:    []string{"z1"},
-			leaderSuccessReadType:   SuccessLeaderRead,
-			followerRegionValid:     true,
-			followerAsyncReload:     Some(false),
-			followerSuccessReplica:  []string{"z1"},
-			followerSuccessReadType: SuccessLeaderRead,
-		},
 		{
 			do:                      leaderServerIsBusy,
 			extra:                   []func(suite *testRegionCacheStaleReadSuite){followerServerIsBusy},
@@ -598,10 +548,11 @@ func testStaleRead(s *testRegionCacheStaleReadSuite, r *RegionCacheTestCase, zon
 		return
 	}
 
+	msg := fmt.Sprintf("%v  %#v", string(resp.Resp.(*kvrpcpb.GetResponse).Value), r)
 	_, successZone, successReadType := s.extractResp(resp)
 	find := false
 	if leaderZone {
-		s.Equal(r.leaderSuccessReadType, successReadType)
+		s.Equal(r.leaderSuccessReadType, successReadType, msg)
 		for _, z := range r.leaderSuccessReplica {
 			if z == successZone {
 				find = true
@@ -617,7 +568,7 @@ func testStaleRead(s *testRegionCacheStaleReadSuite, r *RegionCacheTestCase, zon
 			}
 		}
 	}
-	s.True(find)
+	s.True(find, msg)
 }
 
 type Option[T interface{}] struct {
@@ -767,22 +718,6 @@ func leaderDownAndElect(s *testRegionCacheStaleReadSuite) {
 	s.setUnavailableStore(leader.Id)
 }
 
-func leaderDataIsNotReady(s *testRegionCacheStaleReadSuite) {
-	peerID, _ := s.getLeader()
-	s.injection.leaderRegionError = func(req *tikvrpc.Request, zone string) *errorpb.Error {
-		if !req.StaleRead || zone != "z1" {
-			return nil
-		}
-		return &errorpb.Error{
-			DataIsNotReady: &errorpb.DataIsNotReady{
-				RegionId: s.regionID,
-				PeerId:   peerID,
-				SafeTs:   0,
-			},
-		}
-	}
-}
-
 func leaderServerIsBusy(s *testRegionCacheStaleReadSuite) {
 	s.injection.leaderRegionError = func(req *tikvrpc.Request, zone string) *errorpb.Error {
 		if zone != "z1" {
diff --git a/internal/locate/region_request_test.go b/internal/locate/region_request_test.go
index 58d046a2d1..1d12d1f968 100644
--- a/internal/locate/region_request_test.go
+++ b/internal/locate/region_request_test.go
@@ -679,42 +679,6 @@ func (s *testRegionRequestToSingleStoreSuite) TestCloseConnectionOnStoreNotMatch
 	s.Equal(target, client.closedAddr)
 }
 
-func (s *testRegionRequestToSingleStoreSuite) TestStaleReadRetry() {
-	req := tikvrpc.NewRequest(tikvrpc.CmdGet, &kvrpcpb.GetRequest{
-		Key: []byte("key"),
-	})
-	req.EnableStaleWithMixedReplicaRead()
-	req.ReadReplicaScope = "z1" // not global stale read.
-	region, err := s.cache.LocateRegionByID(s.bo, s.region)
-	s.Nil(err)
-	s.NotNil(region)
-
-	oc := s.regionRequestSender.client
-	defer func() {
-		s.regionRequestSender.client = oc
-	}()
-
-	s.regionRequestSender.client = &fnClient{fn: func(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (response *tikvrpc.Response, err error) {
-		if req.StaleRead {
-			// Mock for stale-read request always return DataIsNotReady error when tikv `ResolvedTS` is blocked.
-			response = &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{
-				RegionError: &errorpb.Error{DataIsNotReady: &errorpb.DataIsNotReady{}},
-			}}
-		} else {
-			response = &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{Value: []byte("value")}}
-		}
-		return response, nil
-	}}
-
-	bo := retry.NewBackofferWithVars(context.Background(), 5, nil)
-	resp, _, err := s.regionRequestSender.SendReq(bo, req, region.Region, time.Second)
-	s.Nil(err)
-	s.NotNil(resp)
-	regionErr, _ := resp.GetRegionError()
-	s.Nil(regionErr)
-	s.Equal([]byte("value"), resp.Resp.(*kvrpcpb.GetResponse).Value)
-}
-
 func (s *testRegionRequestToSingleStoreSuite) TestKVReadTimeoutWithDisableBatchClient() {
 	config.UpdateGlobal(func(conf *config.Config) {
 		conf.TiKVClient.MaxBatchSize = 0
diff --git a/internal/locate/replica_selector_test.go b/internal/locate/replica_selector_test.go
new file mode 100644
index 0000000000..3e1b40caa0
--- /dev/null
+++ b/internal/locate/replica_selector_test.go
@@ -0,0 +1,533 @@
+package locate
+
+import (
+	"context"
+	"fmt"
+	"math/rand"
+	"sort"
+	"strconv"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/pingcap/failpoint"
+	"github.com/pingcap/kvproto/pkg/errorpb"
+	"github.com/pingcap/kvproto/pkg/kvrpcpb"
+	"github.com/pingcap/kvproto/pkg/metapb"
+	"github.com/stretchr/testify/suite"
+	"github.com/tikv/client-go/v2/config/retry"
+	"github.com/tikv/client-go/v2/internal/apicodec"
+	"github.com/tikv/client-go/v2/internal/client"
+	"github.com/tikv/client-go/v2/internal/mockstore/mocktikv"
+	"github.com/tikv/client-go/v2/kv"
+	"github.com/tikv/client-go/v2/oracle"
+	"github.com/tikv/client-go/v2/tikvrpc"
+)
+
+type testReplicaSelectorSuite struct {
+	suite.Suite
+	cluster    *mocktikv.Cluster
+	storeIDs   []uint64
+	peerIDs    []uint64
+	regionID   uint64
+	leaderPeer uint64
+	cache      *RegionCache
+	bo         *retry.Backoffer
+	mvccStore  mocktikv.MVCCStore
+}
+
+func (s *testReplicaSelectorSuite) SetupTest(t *testing.T) {
+	s.mvccStore = mocktikv.MustNewMVCCStore()
+	s.cluster = mocktikv.NewCluster(s.mvccStore)
+	s.storeIDs, s.peerIDs, s.regionID, s.leaderPeer = mocktikv.BootstrapWithMultiStores(s.cluster, 3)
+	pdCli := &CodecPDClient{mocktikv.NewPDClient(s.cluster), apicodec.NewCodecV1(apicodec.ModeTxn)}
+	s.cache = NewRegionCache(pdCli)
+	s.bo = retry.NewNoopBackoff(context.Background())
+	s.SetT(t)
+	s.SetS(s)
+
+	randIntn = func(n int) int { return 0 }
+	s.NoError(failpoint.Enable("tikvclient/fastBackoffBySkipSleep", `return`))
+	s.NoError(failpoint.Enable("tikvclient/skipStoreCheckUntilHealth", `return`))
+
+	loc, err := s.cache.LocateKey(s.bo, []byte("key"))
+	s.Nil(err)
+	r := s.cache.GetCachedRegionWithRLock(loc.Region)
+	s.NotNil(r)
+	// The following assumptions are made in the latter tests, which should be checked in advance:
+	s.Equal(r.GetLeaderStoreID(), uint64(1)) // region's leader in store1.
+	s.Equal(len(r.getStore().stores), 3)     // region has 3 peer(stores).
+	for _, store := range r.getStore().stores {
+		s.Equal(store.labels[0].Key, "id") // Each store has a label "id", and the value is the store's ID.
+		s.Equal(store.labels[0].Value, fmt.Sprintf("%v", store.storeID))
+	}
+}
+
+func (s *testReplicaSelectorSuite) TearDownTest() {
+	s.cache.Close()
+	s.mvccStore.Close()
+
+	randIntn = rand.Intn
+	s.NoError(failpoint.Disable("tikvclient/fastBackoffBySkipSleep"))
+	s.NoError(failpoint.Disable("tikvclient/skipStoreCheckUntilHealth"))
+}
+
+type replicaSelectorAccessPathCase struct {
+	reqType          tikvrpc.CmdType
+	readType         kv.ReplicaReadType
+	staleRead        bool
+	timeout          time.Duration
+	label            *metapb.StoreLabel
+	accessErr        []RegionErrorType
+	accessErrInValid bool
+	accessPathResult                   // use to record the execution result.
+	expect           *accessPathResult //
+}
+
+type accessPathResult struct {
+	accessPath      []string
+	respErr         string
+	respRegionError *errorpb.Error
+	backoffCnt      int
+	backoffDetail   []string
+	regionIsValid   bool
+}
+
+func TestReplicaReadStaleReadAccessPathByCase(t *testing.T) {
+	s := new(testReplicaSelectorSuite)
+	s.SetupTest(t)
+	defer s.TearDownTest()
+
+	var ca replicaSelectorAccessPathCase
+	ca = replicaSelectorAccessPathCase{
+		reqType:   tikvrpc.CmdGet,
+		readType:  kv.ReplicaReadMixed,
+		staleRead: true,
+		accessErr: []RegionErrorType{DataIsNotReadyErr, ServerIsBusyErr},
+		expect: &accessPathResult{
+			accessPath: []string{
+				"{addr: store1, replica-read: false, stale-read: true}",
+				"{addr: store2, replica-read: true, stale-read: false}",
+				"{addr: store3, replica-read: true, stale-read: false}",
+			},
+			respErr:         "",
+			respRegionError: nil,
+			backoffCnt:      1,
+			backoffDetail:   []string{"tikvServerBusy+1"},
+			regionIsValid:   true,
+		},
+	}
+	s.True(s.runCaseAndCompare(ca))
+
+	// test stale read with label.
+	ca = replicaSelectorAccessPathCase{
+		reqType:   tikvrpc.CmdGet,
+		readType:  kv.ReplicaReadMixed,
+		staleRead: true,
+		label:     &metapb.StoreLabel{Key: "id", Value: "2"},
+		accessErr: []RegionErrorType{DataIsNotReadyErr},
+		expect: &accessPathResult{
+			accessPath: []string{
+				"{addr: store2, replica-read: false, stale-read: true}",
+				"{addr: store1, replica-read: false, stale-read: false}", // try leader with leader read.
+			},
+			respErr:         "",
+			respRegionError: nil,
+			backoffCnt:      0,
+			backoffDetail:   []string{},
+			regionIsValid:   true,
+		},
+	}
+	s.True(s.runCaseAndCompare(ca))
+
+	ca = replicaSelectorAccessPathCase{
+		reqType:   tikvrpc.CmdGet,
+		readType:  kv.ReplicaReadMixed,
+		staleRead: true,
+		label:     &metapb.StoreLabel{Key: "id", Value: "2"},
+		accessErr: []RegionErrorType{DataIsNotReadyErr, ServerIsBusyErr},
+		expect: &accessPathResult{
+			accessPath: []string{
+				"{addr: store2, replica-read: false, stale-read: true}",
+				"{addr: store1, replica-read: false, stale-read: false}",
+				"{addr: store2, replica-read: true, stale-read: false}"},
+			respErr:         "",
+			respRegionError: nil,
+			backoffCnt:      0,
+			backoffDetail:   []string{},
+			regionIsValid:   true,
+		},
+	}
+	s.True(s.runCaseAndCompare(ca))
+}
+
+func (s *testReplicaSelectorSuite) runCaseAndCompare(ca2 replicaSelectorAccessPathCase) bool {
+	ca2.run(s)
+	if ca2.accessErrInValid {
+		// the case has been marked as invalid, just ignore it.
+		return false
+	}
+	if ca2.expect != nil {
+		msg := fmt.Sprintf("%v\n\n", ca2.Format())
+		expect := ca2.expect
+		result := ca2.accessPathResult
+		s.Equal(expect.accessPath, result.accessPath, msg)
+		s.Equal(expect.respErr, result.respErr, msg)
+		s.Equal(expect.respRegionError, result.respRegionError, msg)
+		s.Equal(expect.regionIsValid, result.regionIsValid, msg)
+		s.Equal(expect.backoffCnt, result.backoffCnt, msg)
+		s.Equal(expect.backoffDetail, result.backoffDetail, msg)
+	}
+	return true
+}
+
+func (ca *replicaSelectorAccessPathCase) run(s *testReplicaSelectorSuite) {
+	reachable.injectConstantLiveness(s.cache) // inject reachable liveness.
+	msg := ca.Format()
+	access := []string{}
+	fnClient := &fnClient{fn: func(ctx context.Context, addr string, req *tikvrpc.Request, timeout time.Duration) (response *tikvrpc.Response, err error) {
+		idx := len(access)
+		access = append(access, fmt.Sprintf("{addr: %v, replica-read: %v, stale-read: %v}", addr, req.ReplicaRead, req.StaleRead))
+		if idx < len(ca.accessErr) {
+			if !ca.accessErr[idx].Valid(addr, req) {
+				// mark this case is invalid. just ignore this case.
+				ca.accessErrInValid = true
+			} else {
+				loc, err := s.cache.LocateKey(s.bo, []byte("key"))
+				s.Nil(err)
+				rc := s.cache.GetCachedRegionWithRLock(loc.Region)
+				s.NotNil(rc)
+				regionErr, err := ca.genAccessErr(s.cache, rc, ca.accessErr[idx])
+				if regionErr != nil {
+					return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{
+						RegionError: regionErr,
+					}}, nil
+				}
+				if err != nil {
+					return nil, err
+				}
+			}
+		}
+		return &tikvrpc.Response{Resp: &kvrpcpb.GetResponse{
+			Value: []byte("hello world"),
+		}}, nil
+	}}
+	sender := NewRegionRequestSender(s.cache, fnClient)
+	var req *tikvrpc.Request
+	switch ca.reqType {
+	case tikvrpc.CmdGet:
+		req = tikvrpc.NewRequest(tikvrpc.CmdGet, &kvrpcpb.GetRequest{
+			Key: []byte("key"),
+		})
+	case tikvrpc.CmdPrewrite:
+		req = tikvrpc.NewRequest(tikvrpc.CmdPrewrite, &kvrpcpb.PrewriteRequest{})
+	default:
+		s.FailNow("unsupported reqType " + ca.reqType.String())
+	}
+	if ca.staleRead {
+		req.EnableStaleWithMixedReplicaRead()
+		req.ReadReplicaScope = oracle.GlobalTxnScope
+		req.TxnScope = oracle.GlobalTxnScope
+	} else {
+		req.ReplicaReadType = ca.readType
+		req.ReplicaRead = ca.readType.IsFollowerRead()
+	}
+	opts := []StoreSelectorOption{}
+	if ca.label != nil {
+		opts = append(opts, WithMatchLabels([]*metapb.StoreLabel{ca.label}))
+	}
+	// reset slow score, since serverIsBusyErr will mark the store is slow, and affect remaining test cases.
+	loc, err := s.cache.LocateKey(s.bo, []byte("key"))
+	s.Nil(err)
+	rc := s.cache.GetCachedRegionWithRLock(loc.Region)
+	s.NotNil(rc)
+	for _, store := range rc.getStore().stores {
+		store.slowScore.resetSlowScore()
+		atomic.StoreUint32(&store.livenessState, uint32(reachable))
+		store.setResolveState(resolved)
+	}
+
+	bo := retry.NewBackofferWithVars(context.Background(), 40000, nil)
+	timeout := ca.timeout
+	if timeout == 0 {
+		timeout = client.ReadTimeoutShort
+	}
+	resp, _, _, err := sender.SendReqCtx(bo, req, loc.Region, timeout, tikvrpc.TiKV, opts...)
+	if err == nil {
+		s.NotNil(resp, msg)
+		regionErr, err := resp.GetRegionError()
+		s.Nil(err, msg)
+		ca.respRegionError = regionErr
+	} else {
+		ca.respErr = err.Error()
+	}
+	ca.accessPath = access
+	ca.backoffCnt = bo.GetTotalBackoffTimes()
+	detail := make([]string, 0, len(bo.GetBackoffTimes()))
+	for tp, cnt := range bo.GetBackoffTimes() {
+		detail = append(detail, fmt.Sprintf("%v+%v", tp, cnt))
+	}
+	sort.Strings(detail)
+	ca.backoffDetail = detail
+	ca.regionIsValid = sender.replicaSelector.region.isValid()
+	sender.replicaSelector.invalidateRegion() // invalidate region to reload for next test case.
+}
+
+func (ca *replicaSelectorAccessPathCase) genAccessErr(regionCache *RegionCache, r *Region, accessErr RegionErrorType) (regionErr *errorpb.Error, err error) {
+	genNotLeaderErr := func(storeID uint64) *errorpb.Error {
+		var peerInStore *metapb.Peer
+		for _, peer := range r.meta.Peers {
+			if peer.StoreId == storeID {
+				peerInStore = peer
+				break
+			}
+		}
+		return &errorpb.Error{
+			NotLeader: &errorpb.NotLeader{
+				RegionId: r.meta.Id,
+				Leader:   peerInStore,
+			},
+		}
+	}
+	switch accessErr {
+	case NotLeaderWithNewLeader1Err:
+		regionErr = genNotLeaderErr(1)
+	case NotLeaderWithNewLeader2Err:
+		regionErr = genNotLeaderErr(2)
+	case NotLeaderWithNewLeader3Err:
+		regionErr = genNotLeaderErr(3)
+	default:
+		regionErr, err = accessErr.GenError()
+	}
+	if err != nil {
+		// inject unreachable liveness.
+		unreachable.injectConstantLiveness(regionCache)
+	}
+	return regionErr, err
+}
+
+func (c *replicaSelectorAccessPathCase) Format() string {
+	label := ""
+	if c.label != nil {
+		label = fmt.Sprintf("%v->%v", c.label.Key, c.label.Value)
+	}
+	respRegionError := ""
+	if c.respRegionError != nil {
+		respRegionError = c.respRegionError.String()
+	}
+	accessErr := make([]string, len(c.accessErr))
+	for i := range c.accessErr {
+		accessErr[i] = c.accessErr[i].String()
+	}
+	return fmt.Sprintf("{\n"+
+		"\treq: %v\n"+
+		"\tread_type: %v\n"+
+		"\tstale_read: %v\n"+
+		"\ttimeout: %v\n"+
+		"\tlabel: %v\n"+
+		"\taccess_err: %v\n"+
+		"\taccess_path: %v\n"+
+		"\tresp_err: %v\n"+
+		"\tresp_region_err: %v\n"+
+		"\tbackoff_cnt: %v\n"+
+		"\tbackoff_detail: %v\n"+
+		"\tregion_is_valid: %v\n}",
+		c.reqType, c.readType, c.staleRead, c.timeout, label, strings.Join(accessErr, ", "), strings.Join(c.accessPath, ", "),
+		c.respErr, respRegionError, c.backoffCnt, strings.Join(c.backoffDetail, ", "), c.regionIsValid)
+}
+
+type RegionErrorType int
+
+const (
+	NotLeaderErr RegionErrorType = iota + 1
+	NotLeaderWithNewLeader1Err
+	NotLeaderWithNewLeader2Err
+	NotLeaderWithNewLeader3Err
+	RegionNotFoundErr
+	KeyNotInRegionErr
+	EpochNotMatchErr
+	ServerIsBusyErr
+	ServerIsBusyWithEstimatedWaitMsErr
+	StaleCommandErr
+	StoreNotMatchErr
+	RaftEntryTooLargeErr
+	MaxTimestampNotSyncedErr
+	ReadIndexNotReadyErr
+	ProposalInMergingModeErr
+	DataIsNotReadyErr
+	RegionNotInitializedErr
+	DiskFullErr
+	RecoveryInProgressErr
+	FlashbackInProgressErr
+	FlashbackNotPreparedErr
+	IsWitnessErr
+	MismatchPeerIdErr
+	BucketVersionNotMatchErr
+	// following error type is not region error.
+	DeadLineExceededErr
+	RegionErrorTypeMax
+)
+
+func (tp RegionErrorType) GenRegionError() *errorpb.Error {
+	err := &errorpb.Error{}
+	switch tp {
+	case NotLeaderErr:
+		err.NotLeader = &errorpb.NotLeader{}
+	case RegionNotFoundErr:
+		err.RegionNotFound = &errorpb.RegionNotFound{}
+	case KeyNotInRegionErr:
+		err.KeyNotInRegion = &errorpb.KeyNotInRegion{}
+	case EpochNotMatchErr:
+		err.EpochNotMatch = &errorpb.EpochNotMatch{}
+	case ServerIsBusyErr:
+		err.ServerIsBusy = &errorpb.ServerIsBusy{}
+	case ServerIsBusyWithEstimatedWaitMsErr:
+		err.ServerIsBusy = &errorpb.ServerIsBusy{EstimatedWaitMs: 10}
+	case StaleCommandErr:
+		err.StaleCommand = &errorpb.StaleCommand{}
+	case StoreNotMatchErr:
+		err.StoreNotMatch = &errorpb.StoreNotMatch{}
+	case RaftEntryTooLargeErr:
+		err.RaftEntryTooLarge = &errorpb.RaftEntryTooLarge{}
+	case MaxTimestampNotSyncedErr:
+		err.MaxTimestampNotSynced = &errorpb.MaxTimestampNotSynced{}
+	case ReadIndexNotReadyErr:
+		err.ReadIndexNotReady = &errorpb.ReadIndexNotReady{}
+	case ProposalInMergingModeErr:
+		err.ProposalInMergingMode = &errorpb.ProposalInMergingMode{}
+	case DataIsNotReadyErr:
+		err.DataIsNotReady = &errorpb.DataIsNotReady{}
+	case RegionNotInitializedErr:
+		err.RegionNotInitialized = &errorpb.RegionNotInitialized{}
+	case DiskFullErr:
+		err.DiskFull = &errorpb.DiskFull{}
+	case RecoveryInProgressErr:
+		err.RecoveryInProgress = &errorpb.RecoveryInProgress{}
+	case FlashbackInProgressErr:
+		err.FlashbackInProgress = &errorpb.FlashbackInProgress{}
+	case FlashbackNotPreparedErr:
+		err.FlashbackNotPrepared = &errorpb.FlashbackNotPrepared{}
+	case IsWitnessErr:
+		err.IsWitness = &errorpb.IsWitness{}
+	case MismatchPeerIdErr:
+		err.MismatchPeerId = &errorpb.MismatchPeerId{}
+	case BucketVersionNotMatchErr:
+		err.BucketVersionNotMatch = &errorpb.BucketVersionNotMatch{}
+	default:
+		return nil
+	}
+	return err
+}
+
+func (tp RegionErrorType) GenError() (*errorpb.Error, error) {
+	regionErr := tp.GenRegionError()
+	if regionErr != nil {
+		return regionErr, nil
+	}
+	switch tp {
+	case DeadLineExceededErr:
+		return nil, context.DeadlineExceeded
+	}
+	return nil, nil
+}
+
+func (tp RegionErrorType) Valid(addr string, req *tikvrpc.Request) bool {
+	// leader-read.
+	if !req.StaleRead && !req.ReplicaRead {
+		switch tp {
+		case DataIsNotReadyErr:
+			// DataIsNotReadyErr only return when req is a stale read.
+			return false
+		}
+	}
+	// replica-read.
+	if !req.StaleRead && req.ReplicaRead {
+		switch tp {
+		case NotLeaderErr, NotLeaderWithNewLeader1Err, NotLeaderWithNewLeader2Err, NotLeaderWithNewLeader3Err:
+			// NotLeaderErr will not return in replica read.
+			return false
+		case DataIsNotReadyErr:
+			// DataIsNotReadyErr only return when req is a stale read.
+			return false
+		}
+	}
+	// stale-read.
+	if req.StaleRead && !req.ReplicaRead {
+		switch tp {
+		case NotLeaderErr, NotLeaderWithNewLeader1Err, NotLeaderWithNewLeader2Err, NotLeaderWithNewLeader3Err:
+			// NotLeaderErr will not return in stale read.
+			return false
+		}
+	}
+	// store1 can't return a not leader error with new leader in store1.
+	if addr == "store1" && tp == NotLeaderWithNewLeader1Err {
+		return false
+	}
+	// ditto.
+	if addr == "store2" && tp == NotLeaderWithNewLeader2Err {
+		return false
+	}
+	// ditto.
+	if addr == "store3" && tp == NotLeaderWithNewLeader3Err {
+		return false
+	}
+	return true
+}
+
+func (tp RegionErrorType) String() string {
+	switch tp {
+	case NotLeaderErr:
+		return "NotLeaderErr"
+	case NotLeaderWithNewLeader1Err:
+		return "NotLeaderWithNewLeader1Err"
+	case NotLeaderWithNewLeader2Err:
+		return "NotLeaderWithNewLeader2Err"
+	case NotLeaderWithNewLeader3Err:
+		return "NotLeaderWithNewLeader3Err"
+	case RegionNotFoundErr:
+		return "RegionNotFoundErr"
+	case KeyNotInRegionErr:
+		return "KeyNotInRegionErr"
+	case EpochNotMatchErr:
+		return "EpochNotMatchErr"
+	case ServerIsBusyErr:
+		return "ServerIsBusyErr"
+	case ServerIsBusyWithEstimatedWaitMsErr:
+		return "ServerIsBusyWithEstimatedWaitMsErr"
+	case StaleCommandErr:
+		return "StaleCommandErr"
+	case StoreNotMatchErr:
+		return "StoreNotMatchErr"
+	case RaftEntryTooLargeErr:
+		return "RaftEntryTooLargeErr"
+	case MaxTimestampNotSyncedErr:
+		return "MaxTimestampNotSyncedErr"
+	case ReadIndexNotReadyErr:
+		return "ReadIndexNotReadyErr"
+	case ProposalInMergingModeErr:
+		return "ProposalInMergingModeErr"
+	case DataIsNotReadyErr:
+		return "DataIsNotReadyErr"
+	case RegionNotInitializedErr:
+		return "RegionNotInitializedErr"
+	case DiskFullErr:
+		return "DiskFullErr"
+	case RecoveryInProgressErr:
+		return "RecoveryInProgressErr"
+	case FlashbackInProgressErr:
+		return "FlashbackInProgressErr"
+	case FlashbackNotPreparedErr:
+		return "FlashbackNotPreparedErr"
+	case IsWitnessErr:
+		return "IsWitnessErr"
+	case MismatchPeerIdErr:
+		return "MismatchPeerIdErr"
+	case BucketVersionNotMatchErr:
+		return "BucketVersionNotMatchErr"
+	case DeadLineExceededErr:
+		return "DeadLineExceededErr"
+	default:
+		return "unknown_" + strconv.Itoa(int(tp))
+	}
+}
diff --git a/internal/locate/slow_score.go b/internal/locate/slow_score.go
index 562c9e9db7..d67c846b3c 100644
--- a/internal/locate/slow_score.go
+++ b/internal/locate/slow_score.go
@@ -153,6 +153,11 @@ func (ss *SlowScoreStat) markAlreadySlow() {
 	atomic.StoreUint64(&ss.avgScore, slowScoreMax)
 }
 
+// resetSlowScore resets the slow score to 0. It's used for test.
+func (ss *SlowScoreStat) resetSlowScore() {
+	atomic.StoreUint64(&ss.avgScore, 0)
+}
+
 func (ss *SlowScoreStat) isSlow() bool {
 	return ss.getSlowScore() >= slowScoreThreshold
 }