diff --git a/pkg/mcs/scheduling/server/cluster.go b/pkg/mcs/scheduling/server/cluster.go
index 1e502a542199..d106578b5fe5 100644
--- a/pkg/mcs/scheduling/server/cluster.go
+++ b/pkg/mcs/scheduling/server/cluster.go
@@ -568,3 +568,10 @@ func (c *Cluster) processRegionHeartbeat(region *core.RegionInfo) error {
 func (c *Cluster) IsPrepared() bool {
 	return c.coordinator.GetPrepareChecker().IsPrepared()
 }
+
+// IsSchedulingHalted returns whether the scheduling is halted.
+// Currently, the microservice scheduling is halted when:
+//   - The `HaltScheduling` persist option is set to true.
+func (c *Cluster) IsSchedulingHalted() bool {
+	return c.persistConfig.IsSchedulingHalted()
+}
diff --git a/pkg/mcs/scheduling/server/config/config.go b/pkg/mcs/scheduling/server/config/config.go
index d462a9a58b53..72e46cb341fd 100644
--- a/pkg/mcs/scheduling/server/config/config.go
+++ b/pkg/mcs/scheduling/server/config/config.go
@@ -603,6 +603,10 @@ func (o *PersistConfig) SetSplitMergeInterval(splitMergeInterval time.Duration)
 	o.SetScheduleConfig(v)
 }
 
+// SetSchedulingAllowanceStatus sets the scheduling allowance status to help distinguish the source of the halt.
+// TODO: support this metrics for the scheduling service in the future.
+func (*PersistConfig) SetSchedulingAllowanceStatus(bool, string) {}
+
 // SetHaltScheduling set HaltScheduling.
 func (o *PersistConfig) SetHaltScheduling(halt bool, source string) {
 	v := o.GetScheduleConfig().Clone()
diff --git a/pkg/schedule/config/config_provider.go b/pkg/schedule/config/config_provider.go
index 20c7f0dc2cf1..90e489f86f3e 100644
--- a/pkg/schedule/config/config_provider.go
+++ b/pkg/schedule/config/config_provider.go
@@ -46,7 +46,7 @@ func IsSchedulerRegistered(name string) bool {
 type SchedulerConfigProvider interface {
 	SharedConfigProvider
 
-	IsSchedulingHalted() bool
+	SetSchedulingAllowanceStatus(bool, string)
 	GetStoresLimit() map[uint64]StoreLimitConfig
 
 	IsSchedulerDisabled(string) bool
diff --git a/pkg/schedule/coordinator.go b/pkg/schedule/coordinator.go
index fcdd8c9a32c8..7d4b839385ab 100644
--- a/pkg/schedule/coordinator.go
+++ b/pkg/schedule/coordinator.go
@@ -178,7 +178,7 @@ func (c *Coordinator) PatrolRegions() {
 			log.Info("patrol regions has been stopped")
 			return
 		}
-		if c.isSchedulingHalted() {
+		if c.cluster.IsSchedulingHalted() {
 			continue
 		}
 
@@ -207,10 +207,6 @@ func (c *Coordinator) PatrolRegions() {
 	}
 }
 
-func (c *Coordinator) isSchedulingHalted() bool {
-	return c.cluster.GetSchedulerConfig().IsSchedulingHalted()
-}
-
 func (c *Coordinator) checkRegions(startKey []byte) (key []byte, regions []*core.RegionInfo) {
 	regions = c.cluster.ScanRegions(startKey, nil, patrolScanRegionLimit)
 	if len(regions) == 0 {
diff --git a/pkg/schedule/core/cluster_informer.go b/pkg/schedule/core/cluster_informer.go
index 63dacd0c30dd..b97459d26ea6 100644
--- a/pkg/schedule/core/cluster_informer.go
+++ b/pkg/schedule/core/cluster_informer.go
@@ -43,6 +43,7 @@ type SchedulerCluster interface {
 	GetSchedulerConfig() sc.SchedulerConfigProvider
 	GetRegionLabeler() *labeler.RegionLabeler
 	GetStoreConfig() sc.StoreConfigProvider
+	IsSchedulingHalted() bool
 }
 
 // CheckerCluster is an aggregate interface that wraps multiple interfaces
diff --git a/pkg/schedule/schedulers/scheduler_controller.go b/pkg/schedule/schedulers/scheduler_controller.go
index 25a2c8b2afe4..28bb5c96c030 100644
--- a/pkg/schedule/schedulers/scheduler_controller.go
+++ b/pkg/schedule/schedulers/scheduler_controller.go
@@ -114,7 +114,7 @@ func (c *Controller) CollectSchedulerMetrics() {
 		var allowScheduler float64
 		// If the scheduler is not allowed to schedule, it will disappear in Grafana panel.
 		// See issue #1341.
-		if !s.IsPaused() && !c.isSchedulingHalted() {
+		if !s.IsPaused() && !c.cluster.IsSchedulingHalted() {
 			allowScheduler = 1
 		}
 		schedulerStatusGauge.WithLabelValues(s.Scheduler.GetName(), "allow").Set(allowScheduler)
@@ -130,10 +130,6 @@ func (c *Controller) CollectSchedulerMetrics() {
 	ruleStatusGauge.WithLabelValues("group_count").Set(float64(groupCnt))
 }
 
-func (c *Controller) isSchedulingHalted() bool {
-	return c.cluster.GetSchedulerConfig().IsSchedulingHalted()
-}
-
 // ResetSchedulerMetrics resets metrics of all schedulers.
 func ResetSchedulerMetrics() {
 	schedulerStatusGauge.Reset()
@@ -518,7 +514,7 @@ func (s *ScheduleController) AllowSchedule(diagnosable bool) bool {
 		}
 		return false
 	}
-	if s.isSchedulingHalted() {
+	if s.cluster.IsSchedulingHalted() {
 		if diagnosable {
 			s.diagnosticRecorder.SetResultFromStatus(Halted)
 		}
@@ -533,10 +529,6 @@ func (s *ScheduleController) AllowSchedule(diagnosable bool) bool {
 	return true
 }
 
-func (s *ScheduleController) isSchedulingHalted() bool {
-	return s.cluster.GetSchedulerConfig().IsSchedulingHalted()
-}
-
 // IsPaused returns if a scheduler is paused.
 func (s *ScheduleController) IsPaused() bool {
 	delayUntil := atomic.LoadInt64(&s.delayUntil)
diff --git a/pkg/unsaferecovery/unsafe_recovery_controller.go b/pkg/unsaferecovery/unsafe_recovery_controller.go
index aa45ba6a2bdc..3ce299ca6e01 100644
--- a/pkg/unsaferecovery/unsafe_recovery_controller.go
+++ b/pkg/unsaferecovery/unsafe_recovery_controller.go
@@ -492,12 +492,11 @@ func (u *Controller) GetStage() stage {
 }
 
 func (u *Controller) changeStage(stage stage) {
-	u.stage = stage
-	// Halt and resume the scheduling once the running state changed.
-	running := isRunning(stage)
-	if opt := u.cluster.GetSchedulerConfig(); opt.IsSchedulingHalted() != running {
-		opt.SetHaltScheduling(running, "online-unsafe-recovery")
+	// If the running stage changes, update the scheduling allowance status to add or remove "online-unsafe-recovery" halt.
+	if running := isRunning(stage); running != isRunning(u.stage) {
+		u.cluster.GetSchedulerConfig().SetSchedulingAllowanceStatus(running, "online-unsafe-recovery")
 	}
+	u.stage = stage
 
 	var output StageOutput
 	output.Time = time.Now().Format("2006-01-02 15:04:05.000")
diff --git a/server/cluster/cluster.go b/server/cluster/cluster.go
index d6454f48ee36..b15b03f125ee 100644
--- a/server/cluster/cluster.go
+++ b/server/cluster/cluster.go
@@ -920,6 +920,14 @@ func (c *RaftCluster) RemoveSuspectRegion(id uint64) {
 	c.coordinator.GetCheckerController().RemoveSuspectRegion(id)
 }
 
+// IsSchedulingHalted returns whether the scheduling is halted.
+// Currently, the PD scheduling is halted when:
+//   - The `HaltScheduling` persist option is set to true.
+//   - Online unsafe recovery is running.
+func (c *RaftCluster) IsSchedulingHalted() bool {
+	return c.opt.IsSchedulingHalted() || c.unsafeRecoveryController.IsRunning()
+}
+
 // GetUnsafeRecoveryController returns the unsafe recovery controller.
 func (c *RaftCluster) GetUnsafeRecoveryController() *unsaferecovery.Controller {
 	return c.unsafeRecoveryController
diff --git a/server/cluster/cluster_worker.go b/server/cluster/cluster_worker.go
index c1da97363b53..781b3079703b 100644
--- a/server/cluster/cluster_worker.go
+++ b/server/cluster/cluster_worker.go
@@ -43,7 +43,7 @@ func (c *RaftCluster) HandleRegionHeartbeat(region *core.RegionInfo) error {
 
 // HandleAskSplit handles the split request.
 func (c *RaftCluster) HandleAskSplit(request *pdpb.AskSplitRequest) (*pdpb.AskSplitResponse, error) {
-	if c.isSchedulingHalted() {
+	if c.IsSchedulingHalted() {
 		return nil, errs.ErrSchedulingIsHalted.FastGenByArgs()
 	}
 	if !c.opt.IsTikvRegionSplitEnabled() {
@@ -86,10 +86,6 @@ func (c *RaftCluster) HandleAskSplit(request *pdpb.AskSplitRequest) (*pdpb.AskSp
 	return split, nil
 }
 
-func (c *RaftCluster) isSchedulingHalted() bool {
-	return c.opt.IsSchedulingHalted()
-}
-
 // ValidRequestRegion is used to decide if the region is valid.
 func (c *RaftCluster) ValidRequestRegion(reqRegion *metapb.Region) error {
 	startKey := reqRegion.GetStartKey()
@@ -109,7 +105,7 @@ func (c *RaftCluster) ValidRequestRegion(reqRegion *metapb.Region) error {
 
 // HandleAskBatchSplit handles the batch split request.
 func (c *RaftCluster) HandleAskBatchSplit(request *pdpb.AskBatchSplitRequest) (*pdpb.AskBatchSplitResponse, error) {
-	if c.isSchedulingHalted() {
+	if c.IsSchedulingHalted() {
 		return nil, errs.ErrSchedulingIsHalted.FastGenByArgs()
 	}
 	if !c.opt.IsTikvRegionSplitEnabled() {
diff --git a/server/config/persist_options.go b/server/config/persist_options.go
index c0a0ebf5c47a..032e3736a98d 100644
--- a/server/config/persist_options.go
+++ b/server/config/persist_options.go
@@ -973,11 +973,8 @@ func (o *PersistOptions) SetAllStoresLimitTTL(ctx context.Context, client *clien
 
 var haltSchedulingStatus = schedulingAllowanceStatusGauge.WithLabelValues("halt-scheduling")
 
-// SetHaltScheduling set HaltScheduling.
-func (o *PersistOptions) SetHaltScheduling(halt bool, source string) {
-	v := o.GetScheduleConfig().Clone()
-	v.HaltScheduling = halt
-	o.SetScheduleConfig(v)
+// SetSchedulingAllowanceStatus sets the scheduling allowance status to help distinguish the source of the halt.
+func (*PersistOptions) SetSchedulingAllowanceStatus(halt bool, source string) {
 	if halt {
 		haltSchedulingStatus.Set(1)
 		schedulingAllowanceStatusGauge.WithLabelValues(source).Set(1)
@@ -987,6 +984,14 @@ func (o *PersistOptions) SetHaltScheduling(halt bool, source string) {
 	}
 }
 
+// SetHaltScheduling set HaltScheduling.
+func (o *PersistOptions) SetHaltScheduling(halt bool, source string) {
+	v := o.GetScheduleConfig().Clone()
+	v.HaltScheduling = halt
+	o.SetScheduleConfig(v)
+	o.SetSchedulingAllowanceStatus(halt, source)
+}
+
 // IsSchedulingHalted returns if PD scheduling is halted.
 func (o *PersistOptions) IsSchedulingHalted() bool {
 	if o == nil {
diff --git a/server/server.go b/server/server.go
index 1d2c4516abef..5fc9a64ed289 100644
--- a/server/server.go
+++ b/server/server.go
@@ -1016,6 +1016,8 @@ func (s *Server) SetScheduleConfig(cfg sc.ScheduleConfig) error {
 			errs.ZapError(err))
 		return err
 	}
+	// Update the scheduling halt status at the same time.
+	s.persistOptions.SetSchedulingAllowanceStatus(cfg.HaltScheduling, "manually")
 	log.Info("schedule config is updated", zap.Reflect("new", cfg), zap.Reflect("old", old))
 	return nil
 }