diff --git a/executor/tiflash_test.go b/executor/tiflash_test.go
index a99923712f428..7c2ba3c000c8e 100644
--- a/executor/tiflash_test.go
+++ b/executor/tiflash_test.go
@@ -526,9 +526,6 @@ func (s *tiflashTestSuite) TestDispatchTaskRetry(c *C) {
 	c.Assert(failpoint.Enable("github.com/pingcap/tidb/store/mockstore/unistore/mppDispatchTimeout", "3*return(true)"), IsNil)
 	tk.MustQuery("select count(*) from t").Check(testkit.Rows("4"))
 	c.Assert(failpoint.Disable("github.com/pingcap/tidb/store/mockstore/unistore/mppDispatchTimeout"), IsNil)
-	c.Assert(failpoint.Enable("github.com/pingcap/tidb/store/mockstore/unistore/mppConnTimeout", "3*return(true)"), IsNil)
-	tk.MustQuery("select count(*) from t").Check(testkit.Rows("4"))
-	c.Assert(failpoint.Disable("github.com/pingcap/tidb/store/mockstore/unistore/mppConnTimeout"), IsNil)
 }
 
 func (s *tiflashTestSuite) TestCancelMppTasks(c *C) {
diff --git a/kv/mpp.go b/kv/mpp.go
index fd81f2d264804..c3965831ae05b 100644
--- a/kv/mpp.go
+++ b/kv/mpp.go
@@ -15,6 +15,7 @@ package kv
 
 import (
 	"context"
+	"time"
 
 	"github.com/pingcap/kvproto/pkg/mpp"
 )
@@ -76,7 +77,7 @@ type MPPDispatchRequest struct {
 type MPPClient interface {
 	// ConstructMPPTasks schedules task for a plan fragment.
 	// TODO:: This interface will be refined after we support more executors.
-	ConstructMPPTasks(context.Context, *MPPBuildTasksRequest) ([]MPPTaskMeta, error)
+	ConstructMPPTasks(context.Context, *MPPBuildTasksRequest, map[string]time.Time, time.Duration) ([]MPPTaskMeta, error)
 
 	// DispatchMPPTasks dispatches ALL mpp requests at once, and returns an iterator that transfers the data.
 	DispatchMPPTasks(ctx context.Context, vars interface{}, reqs []*MPPDispatchRequest) Response
diff --git a/planner/core/fragment.go b/planner/core/fragment.go
index c4464c1d19f9b..8adcbf127ff26 100644
--- a/planner/core/fragment.go
+++ b/planner/core/fragment.go
@@ -15,6 +15,7 @@ package core
 
 import (
 	"context"
+	"time"
 
 	"github.com/pingcap/errors"
 	"github.com/pingcap/parser/model"
@@ -346,7 +347,12 @@ func (e *mppTaskGenerator) constructMPPTasksImpl(ctx context.Context, ts *Physic
 
 func (e *mppTaskGenerator) constructMPPTasksForSinglePartitionTable(ctx context.Context, kvRanges []kv.KeyRange, tableID int64) ([]*kv.MPPTask, error) {
 	req := &kv.MPPBuildTasksRequest{KeyRanges: kvRanges}
-	metas, err := e.ctx.GetMPPClient().ConstructMPPTasks(ctx, req)
+	ttl, err := time.ParseDuration(e.ctx.GetSessionVars().MPPStoreFailTTL)
+	if err != nil {
+		logutil.BgLogger().Warn("MPP store fail ttl is invalid", zap.Error(err))
+		ttl = 30 * time.Second
+	}
+	metas, err := e.ctx.GetMPPClient().ConstructMPPTasks(ctx, req, e.ctx.GetSessionVars().MPPStoreLastFailTime, ttl)
 	if err != nil {
 		return nil, errors.Trace(err)
 	}
diff --git a/sessionctx/variable/session.go b/sessionctx/variable/session.go
index b8ee35d0fcc40..2325f50fdb46e 100644
--- a/sessionctx/variable/session.go
+++ b/sessionctx/variable/session.go
@@ -855,6 +855,12 @@ type SessionVars struct {
 
 	// EnableStableResultMode if stabilize query results.
 	EnableStableResultMode bool
+
+	// MPPStoreLastFailTime records the lastest fail time that a TiFlash store failed.
+	MPPStoreLastFailTime map[string]time.Time
+
+	// MPPStoreFailTTL indicates the duration that protect TiDB from sending task to a new recovered TiFlash.
+	MPPStoreFailTTL string
 }
 
 // AllocMPPTaskID allocates task id for mpp tasks. It will reset the task id if the query's
@@ -1069,6 +1075,8 @@ func NewSessionVars() *SessionVars {
 		CTEMaxRecursionDepth:        DefCTEMaxRecursionDepth,
 		TMPTableSize:                DefTMPTableSize,
 		EnableGlobalTemporaryTable:  DefTiDBEnableGlobalTemporaryTable,
+		MPPStoreLastFailTime:        make(map[string]time.Time),
+		MPPStoreFailTTL:             DefTiDBMPPStoreFailTTL,
 	}
 	vars.KVVars = tikvstore.NewVariables(&vars.Killed)
 	vars.Concurrency = Concurrency{
@@ -1117,6 +1125,7 @@ func NewSessionVars() *SessionVars {
 	vars.AllowBatchCop = DefTiDBAllowBatchCop
 	vars.allowMPPExecution = DefTiDBAllowMPPExecution
 	vars.enforceMPPExecution = DefTiDBEnforceMPPExecution
+	vars.MPPStoreFailTTL = DefTiDBMPPStoreFailTTL
 
 	var enableChunkRPC string
 	if config.GetGlobalConfig().TiKVClient.EnableChunkRPC {
diff --git a/sessionctx/variable/sysvar.go b/sessionctx/variable/sysvar.go
index 05da31bea6143..8c739f443a727 100644
--- a/sessionctx/variable/sysvar.go
+++ b/sessionctx/variable/sysvar.go
@@ -853,6 +853,10 @@ var defaultSysVars = []*SysVar{
 		s.allowMPPExecution = TiDBOptOn(val)
 		return nil
 	}},
+	{Scope: ScopeGlobal | ScopeSession, Name: TiDBMPPStoreFailTTL, Type: TypeStr, Value: DefTiDBMPPStoreFailTTL, SetSession: func(s *SessionVars, val string) error {
+		s.MPPStoreFailTTL = val
+		return nil
+	}},
 	{Scope: ScopeSession, Name: TiDBEnforceMPPExecution, Type: TypeBool, Value: BoolToOnOff(config.GetGlobalConfig().Performance.EnforceMPP), Validation: func(vars *SessionVars, normalizedValue string, originalValue string, scope ScopeFlag) (string, error) {
 		if TiDBOptOn(normalizedValue) && !vars.allowMPPExecution {
 			return normalizedValue, ErrWrongValueForVar.GenWithStackByArgs("tidb_enforce_mpp", "1' but tidb_allow_mpp is 0, please activate tidb_allow_mpp at first.")
diff --git a/sessionctx/variable/tidb_vars.go b/sessionctx/variable/tidb_vars.go
index 00c445ada3de6..65f6f01751745 100644
--- a/sessionctx/variable/tidb_vars.go
+++ b/sessionctx/variable/tidb_vars.go
@@ -310,6 +310,10 @@ const (
 	// Note if you want to set `tidb_enforce_mpp` to `true`, you must set `tidb_allow_mpp` to `true` first.
 	TiDBEnforceMPPExecution = "tidb_enforce_mpp"
 
+	// TiDBMPPStoreFailTTL is the unavailable time when a store is detected failed. During that time, tidb will not send any task to
+	// TiFlash even though the failed TiFlash node has been recovered.
+	TiDBMPPStoreFailTTL = "tidb_mpp_store_fail_ttl"
+
 	// TiDBInitChunkSize is used to control the init chunk size during query execution.
 	TiDBInitChunkSize = "tidb_init_chunk_size"
 
@@ -656,6 +660,7 @@ const (
 	DefTiDBAllowBatchCop               = 1
 	DefTiDBAllowMPPExecution           = true
 	DefTiDBEnforceMPPExecution         = false
+	DefTiDBMPPStoreFailTTL             = "60s"
 	DefTiDBTxnMode                     = ""
 	DefTiDBRowFormatV1                 = 1
 	DefTiDBRowFormatV2                 = 2
diff --git a/store/copr/batch_coprocessor.go b/store/copr/batch_coprocessor.go
index c110d9cd34751..2b4e5eb5cc597 100644
--- a/store/copr/batch_coprocessor.go
+++ b/store/copr/batch_coprocessor.go
@@ -103,10 +103,11 @@ func (rs *batchCopResponse) RespTime() time.Duration {
 // 2. for the remaining regions:
 //    if there is only 1 available store, then put the region to the related store
 //    otherwise, use a greedy algorithm to put it into the store with highest weight
-func balanceBatchCopTask(ctx context.Context, kvStore *kvStore, originalTasks []*batchCopTask, isMPP bool) []*batchCopTask {
+func balanceBatchCopTask(ctx context.Context, kvStore *kvStore, originalTasks []*batchCopTask, mppStoreLastFailTime map[string]time.Time, ttl time.Duration) []*batchCopTask {
 	if len(originalTasks) <= 1 {
 		return originalTasks
 	}
+	isMPP := mppStoreLastFailTime != nil
 	cache := kvStore.GetRegionCache()
 	storeTaskMap := make(map[uint64]*batchCopTask)
 	// storeCandidateRegionMap stores all the possible store->region map. Its content is
@@ -133,25 +134,39 @@ func balanceBatchCopTask(ctx context.Context, kvStore *kvStore, originalTasks []
 		var wg sync.WaitGroup
 		var mu sync.Mutex
 		wg.Add(len(stores))
+		cur := time.Now()
 		for i := range stores {
 			go func(idx int) {
 				defer wg.Done()
 				s := stores[idx]
-				aliveReq := tikvrpc.NewRequest(tikvrpc.CmdMPPAlive, &mpp.IsAliveRequest{}, kvrpcpb.Context{})
-				aliveReq.StoreTp = tikvrpc.TiFlash
-				alive := false
-				resp, err := kvStore.GetTiKVClient().SendRequest(ctx, s.GetAddr(), aliveReq, 3*time.Second)
-				if err != nil {
+
+				var last time.Time
+				var ok bool
+				mu.Lock()
+				if last, ok = mppStoreLastFailTime[s.GetAddr()]; ok && cur.Sub(last) < 100*time.Millisecond {
+					// The interval time is so short that may happen in a same query, so we needn't to check again.
+					mu.Unlock()
+					return
+				}
+				mu.Unlock()
+
+				resp, err := kvStore.GetTiKVClient().SendRequest(ctx, s.GetAddr(), &tikvrpc.Request{
+					Type:    tikvrpc.CmdMPPAlive,
+					StoreTp: tikvrpc.TiFlash,
+					Req:     &mpp.IsAliveRequest{},
+					Context: kvrpcpb.Context{},
+				}, 2*time.Second)
+
+				if err != nil || !resp.Resp.(*mpp.IsAliveResponse).Available {
 					logutil.BgLogger().Warn("Cannot detect store's availability", zap.String("store address", s.GetAddr()), zap.String("err message", err.Error()))
-				} else {
-					rpcResp := resp.Resp.(*mpp.IsAliveResponse)
-					if rpcResp.Available {
-						alive = true
-					} else {
-						logutil.BgLogger().Warn("Cannot detect store's availability", zap.String("store address", s.GetAddr()))
-					}
+					mu.Lock()
+					mppStoreLastFailTime[s.GetAddr()] = time.Now()
+					mu.Unlock()
+					return
 				}
-				if !alive {
+
+				if cur.Sub(last) < ttl {
+					logutil.BgLogger().Warn("Cannot detect store's availability because the current time has not reached MPPStoreLastFailTime + MPPStoreFailTTL", zap.String("store address", s.GetAddr()), zap.Time("last fail time", last))
 					return
 				}
 
@@ -292,7 +307,7 @@ func balanceBatchCopTask(ctx context.Context, kvStore *kvStore, originalTasks []
 	return ret
 }
 
-func buildBatchCopTasks(bo *Backoffer, store *kvStore, ranges *KeyRanges, storeType kv.StoreType, isMPP bool) ([]*batchCopTask, error) {
+func buildBatchCopTasks(bo *backoff.Backoffer, store *kvStore, ranges *KeyRanges, storeType kv.StoreType, mppStoreLastFailTime map[string]time.Time, ttl time.Duration) ([]*batchCopTask, error) {
 	cache := store.GetRegionCache()
 	start := time.Now()
 	const cmdType = tikvrpc.CmdBatchCop
@@ -367,7 +382,7 @@ func buildBatchCopTasks(bo *Backoffer, store *kvStore, ranges *KeyRanges, storeT
 			}
 			logutil.BgLogger().Debug(msg)
 		}
-		batchTasks = balanceBatchCopTask(bo.GetCtx(), store, batchTasks, isMPP)
+		batchTasks = balanceBatchCopTask(bo.GetCtx(), store, batchTasks, mppStoreLastFailTime, ttl)
 		if log.GetLevel() <= zap.DebugLevel {
 			msg := "After region balance:"
 			for _, task := range batchTasks {
@@ -394,7 +409,7 @@ func (c *CopClient) sendBatch(ctx context.Context, req *kv.Request, vars *tikv.V
 	ctx = context.WithValue(ctx, tikv.TxnStartKey(), req.StartTs)
 	bo := backoff.NewBackofferWithVars(ctx, copBuildTaskMaxBackoff, vars)
 	ranges := NewKeyRanges(req.KeyRanges)
-	tasks, err := buildBatchCopTasks(bo, c.store.kvStore, ranges, req.StoreType, false)
+	tasks, err := buildBatchCopTasks(bo, c.store.kvStore, ranges, req.StoreType, nil, 0)
 	if err != nil {
 		return copErrorResponse{err}
 	}
@@ -528,19 +543,19 @@ func (b *batchCopIterator) handleTask(ctx context.Context, bo *Backoffer, task *
 }
 
 // Merge all ranges and request again.
-func (b *batchCopIterator) retryBatchCopTask(ctx context.Context, bo *Backoffer, batchTask *batchCopTask) ([]*batchCopTask, error) {
+func (b *batchCopIterator) retryBatchCopTask(ctx context.Context, bo *backoff.Backoffer, batchTask *batchCopTask) ([]*batchCopTask, error) {
 	var ranges []kv.KeyRange
 	for _, ri := range batchTask.regionInfos {
 		ri.Ranges.Do(func(ran *kv.KeyRange) {
 			ranges = append(ranges, *ran)
 		})
 	}
-	return buildBatchCopTasks(bo, b.store, NewKeyRanges(ranges), b.req.StoreType, false)
+	return buildBatchCopTasks(bo, b.store, NewKeyRanges(ranges), b.req.StoreType, nil, 0)
 }
 
 const readTimeoutUltraLong = 3600 * time.Second // For requests that may scan many regions for tiflash.
 
-func (b *batchCopIterator) handleTaskOnce(ctx context.Context, bo *Backoffer, task *batchCopTask) ([]*batchCopTask, error) {
+func (b *batchCopIterator) handleTaskOnce(ctx context.Context, bo *backoff.Backoffer, task *batchCopTask) ([]*batchCopTask, error) {
 	sender := NewRegionBatchRequestSender(b.store.GetRegionCache(), b.store.GetTiKVClient())
 	var regionInfos = make([]*coprocessor.RegionInfo, 0, len(task.regionInfos))
 	for _, ri := range task.regionInfos {
diff --git a/store/copr/mpp.go b/store/copr/mpp.go
index ace10b904afee..f192620d82c0c 100644
--- a/store/copr/mpp.go
+++ b/store/copr/mpp.go
@@ -57,14 +57,14 @@ func (c *MPPClient) selectAllTiFlashStore() []kv.MPPTaskMeta {
 }
 
 // ConstructMPPTasks receives ScheduleRequest, which are actually collects of kv ranges. We allocates MPPTaskMeta for them and returns.
-func (c *MPPClient) ConstructMPPTasks(ctx context.Context, req *kv.MPPBuildTasksRequest) ([]kv.MPPTaskMeta, error) {
+func (c *MPPClient) ConstructMPPTasks(ctx context.Context, req *kv.MPPBuildTasksRequest, mppStoreLastFailTime map[string]time.Time, ttl time.Duration) ([]kv.MPPTaskMeta, error) {
 	ctx = context.WithValue(ctx, tikv.TxnStartKey(), req.StartTS)
 	bo := backoff.NewBackofferWithVars(ctx, copBuildTaskMaxBackoff, nil)
 	if req.KeyRanges == nil {
 		return c.selectAllTiFlashStore(), nil
 	}
 	ranges := NewKeyRanges(req.KeyRanges)
-	tasks, err := buildBatchCopTasks(bo, c.store, ranges, kv.TiFlash, true)
+	tasks, err := buildBatchCopTasks(bo, c.store, ranges, kv.TiFlash, mppStoreLastFailTime, ttl)
 	if err != nil {
 		return nil, errors.Trace(err)
 	}
@@ -331,15 +331,9 @@ func (m *mppIterator) establishMPPConns(bo *Backoffer, req *kv.MPPDispatchReques
 	rpcResp, err := m.store.GetTiKVClient().SendRequest(bo.GetCtx(), req.Meta.GetAddress(), wrappedReq, readTimeoutUltraLong)
 
 	if err != nil {
-		logutil.BgLogger().Warn("establish mpp connection meet error, and retrying", zap.String("error", err.Error()), zap.Uint64("timestamp", taskMeta.StartTs), zap.Int64("task", taskMeta.TaskId))
-		err = bo.Backoff(tikv.BoTiFlashRPC(), err)
-		if err != nil {
-			logutil.BgLogger().Warn("establish mpp connection meet error and cannot retry", zap.String("error", err.Error()), zap.Uint64("timestamp", taskMeta.StartTs), zap.Int64("task", taskMeta.TaskId))
-			// we return timeout to trigger tikv's fallback
-			m.sendError(derr.ErrTiFlashServerTimeout)
-			return
-		}
-		m.establishMPPConns(bo, req, taskMeta)
+		logutil.BgLogger().Warn("establish mpp connection meet error and cannot retry", zap.String("error", err.Error()), zap.Uint64("timestamp", taskMeta.StartTs), zap.Int64("task", taskMeta.TaskId))
+		// we return timeout to trigger tikv's fallback
+		m.sendError(derr.ErrTiFlashServerTimeout)
 		return
 	}