diff --git a/components/cluster/command/prune.go b/components/cluster/command/prune.go
index 3faca3fe6f..7c9ae5682b 100644
--- a/components/cluster/command/prune.go
+++ b/components/cluster/command/prune.go
@@ -14,15 +14,6 @@
 package command
 
 import (
-	"fmt"
-
-	"github.com/fatih/color"
-	perrs "github.com/pingcap/errors"
-	"github.com/pingcap/tiup/pkg/cliutil"
-	operator "github.com/pingcap/tiup/pkg/cluster/operation"
-	"github.com/pingcap/tiup/pkg/cluster/spec"
-	"github.com/pingcap/tiup/pkg/cluster/task"
-	"github.com/pingcap/tiup/pkg/logger/log"
 	"github.com/spf13/cobra"
 )
 
@@ -37,68 +28,9 @@ func newPruneCmd() *cobra.Command {
 
 			clusterName := args[0]
 
-			metadata, err := spec.ClusterMetadata(clusterName)
-			if err != nil {
-				return err
-			}
-
-			return destroyTombstoneIfNeed(clusterName, metadata, gOpt, skipConfirm)
+			return manager.DestroyTombstone(clusterName, gOpt, skipConfirm)
 		},
 	}
 
 	return cmd
 }
-
-func destroyTombstoneIfNeed(clusterName string, metadata *spec.ClusterMeta, opt operator.Options, skipConfirm bool) error {
-	topo := metadata.Topology
-
-	if !operator.NeedCheckTombstone(topo) {
-		return nil
-	}
-
-	tlsCfg, err := topo.TLSConfig(tidbSpec.Path(clusterName, spec.TLSCertKeyDir))
-	if err != nil {
-		return perrs.AddStack(err)
-	}
-
-	ctx := task.NewContext()
-	err = ctx.SetSSHKeySet(spec.ClusterPath(clusterName, "ssh", "id_rsa"),
-		spec.ClusterPath(clusterName, "ssh", "id_rsa.pub"))
-	if err != nil {
-		return perrs.AddStack(err)
-	}
-
-	err = ctx.SetClusterSSH(topo, metadata.User, gOpt.SSHTimeout, gOpt.SSHType, topo.BaseTopo().GlobalOptions.SSHType)
-	if err != nil {
-		return perrs.AddStack(err)
-	}
-
-	nodes, err := operator.DestroyTombstone(ctx, topo, true /* returnNodesOnly */, opt, tlsCfg)
-	if err != nil {
-		return perrs.AddStack(err)
-	}
-
-	if len(nodes) == 0 {
-		return nil
-	}
-
-	if !skipConfirm {
-		err = cliutil.PromptForConfirmOrAbortError(
-			color.HiYellowString(fmt.Sprintf("Will destroy these nodes: %v\nDo you confirm this action? [y/N]:", nodes)),
-		)
-		if err != nil {
-			return err
-		}
-	}
-
-	log.Infof("Start destroy Tombstone nodes: %v ...", nodes)
-
-	_, err = operator.DestroyTombstone(ctx, topo, false /* returnNodesOnly */, opt, tlsCfg)
-	if err != nil {
-		return perrs.AddStack(err)
-	}
-
-	log.Infof("Destroy success")
-
-	return spec.SaveClusterMeta(clusterName, metadata)
-}
diff --git a/components/dm/task/update_dm_meta.go b/components/dm/task/update_dm_meta.go
index c2456c4d34..4395d7789a 100644
--- a/components/dm/task/update_dm_meta.go
+++ b/components/dm/task/update_dm_meta.go
@@ -41,50 +41,58 @@ func NewUpdateDMMeta(cluster string, metadata *dmspec.Metadata, deletedNodesID [
 }
 
 // Execute implements the Task interface
+// the metadata especially the topology is in wide use,
+// the other callers point to this field by a pointer,
+// so we should update the original topology directly, and don't make a copy
 func (u *UpdateDMMeta) Execute(ctx *task.Context) error {
-	// make a copy
-	newMeta := &dmspec.Metadata{}
-	*newMeta = *u.metadata
-	newMeta.Topology = &dmspec.Specification{
-		GlobalOptions: u.metadata.Topology.GlobalOptions,
-		// MonitoredOptions: u.metadata.Topology.MonitoredOptions,
-		ServerConfigs: u.metadata.Topology.ServerConfigs,
-	}
-
 	deleted := set.NewStringSet(u.deletedNodesID...)
 	topo := u.metadata.Topology
+	masters := make([]dmspec.MasterSpec, 0)
 	for i, instance := range (&dmspec.DMMasterComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Masters = append(newMeta.Topology.Masters, topo.Masters[i])
+		masters = append(masters, topo.Masters[i])
 	}
+	topo.Masters = masters
+
+	workers := make([]dmspec.WorkerSpec, 0)
 	for i, instance := range (&dmspec.DMWorkerComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Workers = append(newMeta.Topology.Workers, topo.Workers[i])
+		workers = append(workers, topo.Workers[i])
 	}
+	topo.Workers = workers
+
+	monitors := make([]spec.PrometheusSpec, 0)
 	for i, instance := range (&spec.MonitorComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Monitors = append(newMeta.Topology.Monitors, topo.Monitors[i])
+		monitors = append(monitors, topo.Monitors[i])
 	}
+	topo.Monitors = monitors
+
+	grafanas := make([]spec.GrafanaSpec, 0)
 	for i, instance := range (&spec.GrafanaComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Grafanas = append(newMeta.Topology.Grafanas, topo.Grafanas[i])
+		grafanas = append(grafanas, topo.Grafanas[i])
 	}
+	topo.Grafanas = grafanas
+
+	alertmanagers := make([]spec.AlertmanagerSpec, 0)
 	for i, instance := range (&spec.AlertManagerComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Alertmanagers = append(newMeta.Topology.Alertmanagers, topo.Alertmanagers[i])
+		alertmanagers = append(alertmanagers, topo.Alertmanagers[i])
 	}
+	topo.Alertmanagers = alertmanagers
 
-	return dmspec.GetSpecManager().SaveMeta(u.cluster, newMeta)
+	return dmspec.GetSpecManager().SaveMeta(u.cluster, u.metadata)
 }
 
 // Rollback implements the Task interface
diff --git a/pkg/cluster/manager.go b/pkg/cluster/manager.go
index 2afd9ee197..b48968f2ef 100644
--- a/pkg/cluster/manager.go
+++ b/pkg/cluster/manager.go
@@ -1395,53 +1395,51 @@ func (m *Manager) ScaleIn(
 	var regenConfigTasks []task.Task
 	hasImported := false
 	deletedNodes := set.NewStringSet(nodes...)
-	for _, component := range topo.ComponentsByStartOrder() {
-		for _, instance := range component.Instances() {
-			if deletedNodes.Exist(instance.ID()) {
-				continue
-			}
-			deployDir := spec.Abs(base.User, instance.DeployDir())
-			// data dir would be empty for components which don't need it
-			dataDirs := spec.MultiDirAbs(base.User, instance.DataDir())
-			// log dir will always be with values, but might not used by the component
-			logDir := spec.Abs(base.User, instance.LogDir())
+	topo.IterInstance(func(instance spec.Instance) {
+		if deletedNodes.Exist(instance.ID()) {
+			return
+		}
+		deployDir := spec.Abs(base.User, instance.DeployDir())
+		// data dir would be empty for components which don't need it
+		dataDirs := spec.MultiDirAbs(base.User, instance.DataDir())
+		// log dir will always be with values, but might not used by the component
+		logDir := spec.Abs(base.User, instance.LogDir())
 
-			// Download and copy the latest component to remote if the cluster is imported from Ansible
-			tb := task.NewBuilder()
-			if instance.IsImported() {
-				switch compName := instance.ComponentName(); compName {
-				case spec.ComponentGrafana, spec.ComponentPrometheus, spec.ComponentAlertmanager:
-					version := m.bindVersion(compName, base.Version)
-					tb.Download(compName, instance.OS(), instance.Arch(), version).
-						CopyComponent(
-							compName,
-							instance.OS(),
-							instance.Arch(),
-							version,
-							"", // use default srcPath
-							instance.GetHost(),
-							deployDir,
-						)
-				}
-				hasImported = true
+		// Download and copy the latest component to remote if the cluster is imported from Ansible
+		tb := task.NewBuilder()
+		if instance.IsImported() {
+			switch compName := instance.ComponentName(); compName {
+			case spec.ComponentGrafana, spec.ComponentPrometheus, spec.ComponentAlertmanager:
+				version := m.bindVersion(compName, base.Version)
+				tb.Download(compName, instance.OS(), instance.Arch(), version).
+					CopyComponent(
+						compName,
+						instance.OS(),
+						instance.Arch(),
+						version,
+						"", // use default srcPath
+						instance.GetHost(),
+						deployDir,
+					)
 			}
-
-			t := tb.InitConfig(clusterName,
-				base.Version,
-				m.specManager,
-				instance,
-				base.User,
-				true, // always ignore config check result in scale in
-				meta.DirPaths{
-					Deploy: deployDir,
-					Data:   dataDirs,
-					Log:    logDir,
-					Cache:  m.specManager.Path(clusterName, spec.TempConfigPath),
-				},
-			).Build()
-			regenConfigTasks = append(regenConfigTasks, t)
+			hasImported = true
 		}
-	}
+
+		t := tb.InitConfig(clusterName,
+			base.Version,
+			m.specManager,
+			instance,
+			base.User,
+			true, // always ignore config check result in scale in
+			meta.DirPaths{
+				Deploy: deployDir,
+				Data:   dataDirs,
+				Log:    logDir,
+				Cache:  m.specManager.Path(clusterName, spec.TempConfigPath),
+			},
+		).Build()
+		regenConfigTasks = append(regenConfigTasks, t)
+	})
 
 	// handle dir scheme changes
 	if hasImported {
@@ -1461,7 +1459,6 @@ func (m *Manager) ScaleIn(
 			m.specManager.Path(clusterName, "ssh", "id_rsa.pub")).
 		ClusterSSH(topo, base.User, sshTimeout, sshType, metadata.GetTopology().BaseTopo().GlobalOptions.SSHType)
 
-	// TODO: support command scale in operation.
 	scale(b, metadata, tlsCfg)
 
 	t := b.Parallel(force, regenConfigTasks...).Parallel(force, buildDynReloadProm(metadata.GetTopology())...).Build()
@@ -1595,6 +1592,126 @@ func (m *Manager) ScaleOut(
 	return nil
 }
 
+// DestroyTombstone destory and remove instances that is in tombstone state
+func (m *Manager) DestroyTombstone(
+	clusterName string,
+	gOpt operator.Options,
+	skipConfirm bool,
+) error {
+	var (
+		sshTimeout = gOpt.SSHTimeout
+		sshType    = gOpt.SSHType
+	)
+
+	metadata, err := m.meta(clusterName)
+	// allow specific validation errors so that user can recover a broken
+	// cluster if it is somehow in a bad state.
+	if err != nil &&
+		!errors.Is(perrs.Cause(err), spec.ErrNoTiSparkMaster) {
+		return perrs.AddStack(err)
+	}
+
+	topo := metadata.GetTopology()
+	base := metadata.GetBaseMeta()
+
+	clusterMeta := metadata.(*spec.ClusterMeta)
+	cluster := clusterMeta.Topology
+
+	if !operator.NeedCheckTombstone(cluster) {
+		return nil
+	}
+
+	tlsCfg, err := topo.TLSConfig(m.specManager.Path(clusterName, spec.TLSCertKeyDir))
+	if err != nil {
+		return err
+	}
+
+	b := task.NewBuilder().
+		SSHKeySet(
+			m.specManager.Path(clusterName, "ssh", "id_rsa"),
+			m.specManager.Path(clusterName, "ssh", "id_rsa.pub")).
+		ClusterSSH(topo, base.User, sshTimeout, sshType, metadata.GetTopology().BaseTopo().GlobalOptions.SSHType)
+
+	var nodes []string
+	b.
+		Func("FindTomestoneNodes", func(ctx *task.Context) (err error) {
+			nodes, err = operator.DestroyTombstone(ctx, cluster, true /* returnNodesOnly */, gOpt, tlsCfg)
+			if !skipConfirm {
+				err = cliutil.PromptForConfirmOrAbortError(
+					color.HiYellowString(fmt.Sprintf("Will destroy these nodes: %v\nDo you confirm this action? [y/N]:", nodes)),
+				)
+				if err != nil {
+					return err
+				}
+			}
+			log.Infof("Start destroy Tombstone nodes: %v ...", nodes)
+			return err
+		}).
+		ClusterOperate(cluster, operator.DestroyTombstoneOperation, gOpt, tlsCfg).
+		UpdateMeta(clusterName, clusterMeta, nodes).
+		UpdateTopology(clusterName, m.specManager.Path(clusterName), clusterMeta, nodes)
+
+	var regenConfigTasks []task.Task
+	deletedNodes := set.NewStringSet(nodes...)
+	topo.IterInstance(func(instance spec.Instance) {
+		if deletedNodes.Exist(instance.ID()) {
+			return
+		}
+		deployDir := spec.Abs(base.User, instance.DeployDir())
+		// data dir would be empty for components which don't need it
+		dataDirs := spec.MultiDirAbs(base.User, instance.DataDir())
+		// log dir will always be with values, but might not used by the component
+		logDir := spec.Abs(base.User, instance.LogDir())
+
+		// Download and copy the latest component to remote if the cluster is imported from Ansible
+		tb := task.NewBuilder()
+		if instance.IsImported() {
+			switch compName := instance.ComponentName(); compName {
+			case spec.ComponentGrafana, spec.ComponentPrometheus, spec.ComponentAlertmanager:
+				version := m.bindVersion(compName, base.Version)
+				tb.Download(compName, instance.OS(), instance.Arch(), version).
+					CopyComponent(
+						compName,
+						instance.OS(),
+						instance.Arch(),
+						version,
+						"", // use default srcPath
+						instance.GetHost(),
+						deployDir,
+					)
+			}
+		}
+
+		t := tb.InitConfig(clusterName,
+			base.Version,
+			m.specManager,
+			instance,
+			base.User,
+			true, // always ignore config check result in scale in
+			meta.DirPaths{
+				Deploy: deployDir,
+				Data:   dataDirs,
+				Log:    logDir,
+				Cache:  m.specManager.Path(clusterName, spec.TempConfigPath),
+			},
+		).Build()
+		regenConfigTasks = append(regenConfigTasks, t)
+	})
+
+	t := b.Parallel(true, regenConfigTasks...).Parallel(true, buildDynReloadProm(metadata.GetTopology())...).Build()
+	if err := t.Execute(task.NewContext()); err != nil {
+		if errorx.Cast(err) != nil {
+			// FIXME: Map possible task errors and give suggestions.
+			return err
+		}
+		return perrs.Trace(err)
+	}
+
+	log.Infof("Destroy success")
+
+	return nil
+}
+
 func (m *Manager) meta(name string) (metadata spec.Metadata, err error) {
 	exist, err := m.specManager.Exist(name)
 	if err != nil {
diff --git a/pkg/cluster/operation/scale_in.go b/pkg/cluster/operation/scale_in.go
index 4b14c0ca1a..398ad20a3d 100644
--- a/pkg/cluster/operation/scale_in.go
+++ b/pkg/cluster/operation/scale_in.go
@@ -252,16 +252,6 @@ func ScaleInCluster(
 		}
 	}
 
-	pdServers := make([]spec.PDSpec, 0, len(cluster.PDServers))
-	for i := 0; i < len(cluster.PDServers); i++ {
-		s := cluster.PDServers[i]
-		id := s.Host + ":" + strconv.Itoa(s.ClientPort)
-		if !deletedNodes.Exist(id) {
-			pdServers = append(pdServers, s)
-		}
-	}
-	cluster.PDServers = pdServers
-
 	for i := 0; i < len(cluster.TiKVServers); i++ {
 		s := cluster.TiKVServers[i]
 		id := s.Host + ":" + strconv.Itoa(s.Port)
diff --git a/pkg/cluster/task/update_meta.go b/pkg/cluster/task/update_meta.go
index 8e6c1f5caa..8fe272d1c7 100644
--- a/pkg/cluster/task/update_meta.go
+++ b/pkg/cluster/task/update_meta.go
@@ -29,91 +29,122 @@ type UpdateMeta struct {
 }
 
 // Execute implements the Task interface
+// the metadata especially the topology is in wide use,
+// the other callers point to this field by a pointer,
+// so we should update the original topology directly, and don't make a copy
 func (u *UpdateMeta) Execute(ctx *Context) error {
-	// make a copy
-	newMeta := &spec.ClusterMeta{}
-	*newMeta = *u.metadata
-	newMeta.Topology = &spec.Specification{
-		GlobalOptions:    u.metadata.Topology.GlobalOptions,
-		MonitoredOptions: u.metadata.Topology.MonitoredOptions,
-		ServerConfigs:    u.metadata.Topology.ServerConfigs,
-	}
-
 	deleted := set.NewStringSet(u.deletedNodesID...)
 	topo := u.metadata.Topology
+
+	tidbServers := make([]spec.TiDBSpec, 0)
 	for i, instance := range (&spec.TiDBComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.TiDBServers = append(newMeta.Topology.TiDBServers, topo.TiDBServers[i])
+		tidbServers = append(tidbServers, topo.TiDBServers[i])
 	}
+	topo.TiDBServers = tidbServers
+
+	tikvServers := make([]spec.TiKVSpec, 0)
 	for i, instance := range (&spec.TiKVComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.TiKVServers = append(newMeta.Topology.TiKVServers, topo.TiKVServers[i])
+		tikvServers = append(tikvServers, topo.TiKVServers[i])
 	}
+	topo.TiKVServers = tikvServers
+
+	pdServers := make([]spec.PDSpec, 0)
 	for i, instance := range (&spec.PDComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.PDServers = append(newMeta.Topology.PDServers, topo.PDServers[i])
+		pdServers = append(pdServers, topo.PDServers[i])
 	}
+	topo.PDServers = pdServers
+
+	tiflashServers := make([]spec.TiFlashSpec, 0)
 	for i, instance := range (&spec.TiFlashComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.TiFlashServers = append(newMeta.Topology.TiFlashServers, topo.TiFlashServers[i])
+		tiflashServers = append(tiflashServers, topo.TiFlashServers[i])
 	}
+	topo.TiFlashServers = tiflashServers
+
+	pumpServers := make([]spec.PumpSpec, 0)
 	for i, instance := range (&spec.PumpComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.PumpServers = append(newMeta.Topology.PumpServers, topo.PumpServers[i])
+		pumpServers = append(pumpServers, topo.PumpServers[i])
 	}
+	topo.PumpServers = pumpServers
+
+	drainerServers := make([]spec.DrainerSpec, 0)
 	for i, instance := range (&spec.DrainerComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Drainers = append(newMeta.Topology.Drainers, topo.Drainers[i])
+		drainerServers = append(drainerServers, topo.Drainers[i])
 	}
+	topo.Drainers = drainerServers
+
+	cdcServers := make([]spec.CDCSpec, 0)
 	for i, instance := range (&spec.CDCComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.CDCServers = append(newMeta.Topology.CDCServers, topo.CDCServers[i])
+		cdcServers = append(cdcServers, topo.CDCServers[i])
 	}
+	topo.CDCServers = cdcServers
+
+	tisparkWorkers := make([]spec.TiSparkWorkerSpec, 0)
 	for i, instance := range (&spec.TiSparkWorkerComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.TiSparkWorkers = append(newMeta.Topology.TiSparkWorkers, topo.TiSparkWorkers[i])
+		tisparkWorkers = append(tisparkWorkers, topo.TiSparkWorkers[i])
 	}
+	topo.TiSparkWorkers = tisparkWorkers
+
+	tisparkMasters := make([]spec.TiSparkMasterSpec, 0)
 	for i, instance := range (&spec.TiSparkMasterComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.TiSparkMasters = append(newMeta.Topology.TiSparkMasters, topo.TiSparkMasters[i])
+		tisparkMasters = append(tisparkMasters, topo.TiSparkMasters[i])
 	}
+	topo.TiSparkMasters = tisparkMasters
+
+	monitors := make([]spec.PrometheusSpec, 0)
 	for i, instance := range (&spec.MonitorComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Monitors = append(newMeta.Topology.Monitors, topo.Monitors[i])
+		monitors = append(monitors, topo.Monitors[i])
 	}
+	topo.Monitors = monitors
+
+	grafanas := make([]spec.GrafanaSpec, 0)
 	for i, instance := range (&spec.GrafanaComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Grafanas = append(newMeta.Topology.Grafanas, topo.Grafanas[i])
+		grafanas = append(grafanas, topo.Grafanas[i])
 	}
+	topo.Grafanas = grafanas
+
+	alertmanagers := make([]spec.AlertmanagerSpec, 0)
 	for i, instance := range (&spec.AlertManagerComponent{Topology: topo}).Instances() {
 		if deleted.Exist(instance.ID()) {
 			continue
 		}
-		newMeta.Topology.Alertmanagers = append(newMeta.Topology.Alertmanagers, topo.Alertmanagers[i])
+		alertmanagers = append(alertmanagers, topo.Alertmanagers[i])
 	}
-	return spec.SaveClusterMeta(u.cluster, newMeta)
+	topo.Alertmanagers = alertmanagers
+
+	return spec.SaveClusterMeta(u.cluster, u.metadata)
 }
 
 // Rollback implements the Task interface
diff --git a/tests/tiup-cluster/script/scale_core.sh b/tests/tiup-cluster/script/scale_core.sh
index d3f18f7088..508b58a635 100755
--- a/tests/tiup-cluster/script/scale_core.sh
+++ b/tests/tiup-cluster/script/scale_core.sh
@@ -40,12 +40,17 @@ function scale_core() {
     echo "start scale in tidb"
     tiup-cluster $client --yes scale-in $name -N n1:4000
     wait_instance_num_reach $name $total_sub_one $native_ssh
+    # ensure Prometheus's configuration is updated automatically
+    ! tiup-cluster $client exec $name -N n1 --command "grep -q n1:10080 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
+
     echo "start scale out tidb"
     topo=./topo/full_scale_in_tidb.yaml
     tiup-cluster $client --yes scale-out $name $topo
     # after scale-out, ensure the service is enabled
     tiup-cluster $client exec $name -N n1 --command "systemctl status tidb-4000 | grep Loaded |grep 'enabled; vendor'"
+    tiup-cluster $client exec $name -N n1 --command "grep -q n1:10080 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
+    # scale in tikv maybe exists in several minutes or hours, and the GitHub CI is not guaranteed
     # echo "start scale in tikv"
     # tiup-cluster --yes scale-in $name -N n3:20160
     # wait_instance_num_reach $name $total_sub_one $native_ssh
@@ -53,6 +58,19 @@ function scale_core() {
     # topo=./topo/full_scale_in_tikv.yaml
     # tiup-cluster --yes scale-out $name $topo
 
+    echo "start scale in pump"
+    tiup-cluster $client --yes scale-in $name -N n3:8250
+    wait_instance_num_reach $name $total_sub_one $native_ssh
+
+    # ensure Prometheus's configuration is updated automatically
+    ! tiup-cluster $client exec $name -N n1 --command "grep -q n3:8250 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
+
+    echo "start scale out pump"
+    topo=./topo/full_scale_in_pump.yaml
+    tiup-cluster $client --yes scale-out $name $topo
+    # after scale-out, ensure this instance come back
+    tiup-cluster $client exec $name -N n1 --command "grep -q n3:8250 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
+
     echo "start scale in pd"
     tiup-cluster $client --yes scale-in $name -N n3:2379
     wait_instance_num_reach $name $total_sub_one $native_ssh
@@ -60,12 +78,15 @@ function scale_core() {
     # validate https://github.com/pingcap/tiup/issues/786
     # ensure that this instance is removed from the startup scripts of other components that need to rely on PD
     ! tiup-cluster $client exec $name -N n1 --command "grep -q n3:2379 /home/tidb/deploy/tidb-4000/scripts/run_tidb.sh"
+    # ensure Prometheus's configuration is updated automatically
+    ! tiup-cluster $client exec $name -N n1 --command "grep -q n3:2379 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
     echo "start scale out pd"
     topo=./topo/full_scale_in_pd.yaml
     tiup-cluster $client --yes scale-out $name $topo
     # after scale-out, ensure this instance come back
     tiup-cluster $client exec $name -N n1 --command "grep -q n3:2379 /home/tidb/deploy/tidb-4000/scripts/run_tidb.sh"
+    tiup-cluster $client exec $name -N n1 --command "grep -q n3:2379 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
     echo "start scale in tidb"
     tiup-cluster $client --yes scale-in $name -N n2:4000
diff --git a/tests/tiup-dm/test_cmd.sh b/tests/tiup-dm/test_cmd.sh
index d8bf0c9161..4912ea2da7 100755
--- a/tests/tiup-dm/test_cmd.sh
+++ b/tests/tiup-dm/test_cmd.sh
@@ -54,20 +54,26 @@ total_sub_one=12
 echo "start scale in dm-master"
 tiup-dm --yes scale-in $name -N $ipprefix.101:8261
 wait_instance_num_reach $name $total_sub_one false
+# ensure Prometheus's configuration is updated automatically
+! tiup-dm exec $name -N $ipprefix.101 --command "grep -q $ipprefix.101:8261 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
 echo "start scale out dm-master"
 topo_master=./topo/full_scale_in_dm-master.yaml
 sed "s/__IPPREFIX__/$ipprefix/g" $topo_master.tpl > $topo_master
 tiup-dm --yes scale-out $name $topo_master
+tiup-dm exec $name -N $ipprefix.101 --command "grep -q $ipprefix.101:8261 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
 echo "start scale in dm-worker"
 yes | tiup-dm scale-in $name -N $ipprefix.102:8262
 wait_instance_num_reach $name $total_sub_one
+# ensure Prometheus's configuration is updated automatically
+! tiup-dm exec $name -N $ipprefix.101 --command "grep -q $ipprefix.102:8262 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
 echo "start scale out dm-worker"
 topo_worker=./topo/full_scale_in_dm-worker.yaml
 sed "s/__IPPREFIX__/$ipprefix/g" $topo_worker.tpl > $topo_worker
 yes | tiup-dm scale-out $name $topo_worker
+tiup-dm exec $name -N $ipprefix.101 --command "grep -q $ipprefix.102:8262 /home/tidb/deploy/prometheus-9090/conf/prometheus.yml"
 
 echo "start scale in grafana"
 yes | tiup-dm scale-in $name -N $ipprefix.101:3000