diff --git a/pkg/mcs/resource_manager/server/token_bukets.go b/pkg/mcs/resource_manager/server/token_bukets.go
index 5efab52fe68..47608203847 100644
--- a/pkg/mcs/resource_manager/server/token_bukets.go
+++ b/pkg/mcs/resource_manager/server/token_bukets.go
@@ -20,6 +20,8 @@ import (
 
 	"github.com/gogo/protobuf/proto"
 	rmpb "github.com/pingcap/kvproto/pkg/resource_manager"
+	"github.com/pingcap/log"
+	"go.uber.org/zap"
 )
 
 const (
@@ -31,6 +33,7 @@ const (
 	defaultReserveRatio    = 0.5
 	defaultLoanCoefficient = 2
 	maxAssignTokens        = math.MaxFloat64 / 1024 // assume max client connect is 1024
+	slotExpireTimeout      = 10 * time.Minute
 )
 
 // GroupTokenBucket is a token bucket for a resource group.
@@ -62,6 +65,7 @@ type TokenSlot struct {
 	// tokenCapacity is the number of tokens in the slot.
 	tokenCapacity     float64
 	lastTokenCapacity float64
+	lastReqTime       time.Time
 }
 
 // GroupTokenBucketState is the running state of TokenBucket.
@@ -75,7 +79,8 @@ type GroupTokenBucketState struct {
 	LastUpdate  *time.Time `json:"last_update,omitempty"`
 	Initialized bool       `json:"initialized"`
 	// settingChanged is used to avoid that the number of tokens returned is jitter because of changing fill rate.
-	settingChanged bool
+	settingChanged      bool
+	lastCheckExpireSlot time.Time
 }
 
 // Clone returns the copy of GroupTokenBucketState
@@ -95,6 +100,7 @@ func (gts *GroupTokenBucketState) Clone() *GroupTokenBucketState {
 		Initialized:                gts.Initialized,
 		tokenSlots:                 tokenSlots,
 		clientConsumptionTokensSum: gts.clientConsumptionTokensSum,
+		lastCheckExpireSlot:        gts.lastCheckExpireSlot,
 	}
 }
 
@@ -119,16 +125,18 @@ func (gts *GroupTokenBucketState) balanceSlotTokens(
 	clientUniqueID uint64,
 	settings *rmpb.TokenLimitSettings,
 	requiredToken, elapseTokens float64) {
+	now := time.Now()
 	slot, exist := gts.tokenSlots[clientUniqueID]
 	if !exist {
 		// Only slots that require a positive number will be considered alive,
 		// but still need to allocate the elapsed tokens as well.
 		if requiredToken != 0 {
-			slot = &TokenSlot{}
+			slot = &TokenSlot{lastReqTime: now}
 			gts.tokenSlots[clientUniqueID] = slot
 			gts.clientConsumptionTokensSum = 0
 		}
 	} else {
+		slot.lastReqTime = now
 		if gts.clientConsumptionTokensSum >= maxAssignTokens {
 			gts.clientConsumptionTokensSum = 0
 		}
@@ -139,6 +147,16 @@ func (gts *GroupTokenBucketState) balanceSlotTokens(
 		}
 	}
 
+	if time.Since(gts.lastCheckExpireSlot) >= slotExpireTimeout {
+		gts.lastCheckExpireSlot = now
+		for clientUniqueID, slot := range gts.tokenSlots {
+			if time.Since(slot.lastReqTime) >= slotExpireTimeout {
+				delete(gts.tokenSlots, clientUniqueID)
+				log.Info("delete resource group slot because expire", zap.Time("last-req-time", slot.lastReqTime),
+					zap.Any("expire timeout", slotExpireTimeout), zap.Any("del client id", clientUniqueID), zap.Any("len", len(gts.tokenSlots)))
+			}
+		}
+	}
 	if len(gts.tokenSlots) == 0 {
 		return
 	}
@@ -264,6 +282,7 @@ func (gtb *GroupTokenBucket) init(now time.Time, clientID uint64) {
 		lastTokenCapacity: gtb.Tokens,
 	}
 	gtb.LastUpdate = &now
+	gtb.lastCheckExpireSlot = now
 	gtb.Initialized = true
 }