diff --git a/pkg/cluster/operation/upgrade.go b/pkg/cluster/operation/upgrade.go
index 5c4c420858..3cace668ff 100644
--- a/pkg/cluster/operation/upgrade.go
+++ b/pkg/cluster/operation/upgrade.go
@@ -131,17 +131,19 @@ func Upgrade(
 			if instance.IgnoreMonitorAgent() {
 				noAgentHosts.Insert(instance.GetManageHost())
 			}
+
+			// Usage within the switch statement
 			switch component.Name() {
-			case spec.ComponentPD:
-				// defer PD leader to be upgraded after others
-				isLeader, err := instance.(*spec.PDInstance).IsLeader(ctx, topo, int(options.APITimeout), tlsCfg)
+			case spec.ComponentPD, spec.ComponentTSO, spec.ComponentScheduling:
+				// defer PD related leader/primary to be upgraded after others
+				isLeader, err := checkAndDeferPDLeader(ctx, topo, int(options.APITimeout), tlsCfg, instance)
 				if err != nil {
-					logger.Warnf("cannot found pd leader, ignore: %s", err)
+					logger.Warnf("cannot found pd related leader/primary, ignore: %s, instance: %s", err, instance.ID())
 					return err
 				}
 				if isLeader {
 					deferInstances = append(deferInstances, instance)
-					logger.Debugf("Deferred upgrading of PD leader %s", instance.ID())
+					logger.Debugf("Upgrading deferred instance %s...", instance.ID())
 					continue
 				}
 			case spec.ComponentCDC:
@@ -218,6 +220,22 @@ func Upgrade(
 	return RestartMonitored(ctx, uniqueHosts.Slice(), noAgentHosts, topo.GetMonitoredOptions(), options.OptTimeout, systemdMode)
 }
 
+// checkAndDeferPDLeader checks the PD related leader/primary instance's status and defers its upgrade if necessary.
+func checkAndDeferPDLeader(ctx context.Context, topo spec.Topology, apiTimeout int, tlsCfg *tls.Config, instance spec.Instance) (isLeader bool, err error) {
+	switch instance.ComponentName() {
+	case spec.ComponentPD:
+		isLeader, err = instance.(*spec.PDInstance).IsLeader(ctx, topo, apiTimeout, tlsCfg)
+	case spec.ComponentScheduling:
+		isLeader, err = instance.(*spec.SchedulingInstance).IsPrimary(ctx, topo, tlsCfg)
+	case spec.ComponentTSO:
+		isLeader, err = instance.(*spec.TSOInstance).IsPrimary(ctx, topo, tlsCfg)
+	}
+	if err != nil {
+		return false, err
+	}
+	return isLeader, nil
+}
+
 func upgradeInstance(
 	ctx context.Context,
 	topo spec.Topology,
diff --git a/pkg/cluster/spec/scheduling.go b/pkg/cluster/spec/scheduling.go
index 3da3975f2d..31957570d3 100644
--- a/pkg/cluster/spec/scheduling.go
+++ b/pkg/cluster/spec/scheduling.go
@@ -21,6 +21,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/pingcap/errors"
 	"github.com/pingcap/tiup/pkg/cluster/api"
 	"github.com/pingcap/tiup/pkg/cluster/ctxt"
 	"github.com/pingcap/tiup/pkg/cluster/template/scripts"
@@ -29,6 +30,8 @@ import (
 	"github.com/pingcap/tiup/pkg/utils"
 )
 
+var schedulingService = "scheduling"
+
 // SchedulingSpec represents the scheduling topology specification in topology.yaml
 type SchedulingSpec struct {
 	Host                string `yaml:"host"`
@@ -66,7 +69,7 @@ func (s *SchedulingSpec) Status(ctx context.Context, timeout time.Duration, tlsC
 		return "Down"
 	}
 
-	primary, err := pc.GetServicePrimary("scheduling")
+	primary, err := pc.GetServicePrimary(schedulingService)
 	if err != nil {
 		return "ERR"
 	}
@@ -309,6 +312,27 @@ func (i *SchedulingInstance) setTLSConfig(ctx context.Context, enableTLS bool, c
 	return configs, nil
 }
 
+// IsPrimary checks if the instance is primary
+func (i *SchedulingInstance) IsPrimary(ctx context.Context, topo Topology, tlsCfg *tls.Config) (bool, error) {
+	tidbTopo, ok := topo.(*Specification)
+	if !ok {
+		panic("topo should be type of tidb topology")
+	}
+	pdClient := api.NewPDClient(ctx, tidbTopo.GetPDListWithManageHost(), time.Second*5, tlsCfg)
+	primary, err := pdClient.GetServicePrimary(schedulingService)
+	if err != nil {
+		return false, errors.Annotatef(err, "failed to get Scheduling primary %s", i.GetHost())
+	}
+
+	spec := i.InstanceSpec.(*SchedulingSpec)
+	enableTLS := false
+	if tlsCfg != nil {
+		enableTLS = true
+	}
+
+	return primary == spec.GetAdvertiseListenURL(enableTLS), nil
+}
+
 // ScaleConfig deploy temporary config on scaling
 func (i *SchedulingInstance) ScaleConfig(
 	ctx context.Context,
diff --git a/pkg/cluster/spec/tso.go b/pkg/cluster/spec/tso.go
index e309f1d652..3b02f0630e 100644
--- a/pkg/cluster/spec/tso.go
+++ b/pkg/cluster/spec/tso.go
@@ -21,6 +21,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/pingcap/errors"
 	"github.com/pingcap/tiup/pkg/cluster/api"
 	"github.com/pingcap/tiup/pkg/cluster/ctxt"
 	"github.com/pingcap/tiup/pkg/cluster/template/scripts"
@@ -29,6 +30,8 @@ import (
 	"github.com/pingcap/tiup/pkg/utils"
 )
 
+var tsoService = "tso"
+
 // TSOSpec represents the TSO topology specification in topology.yaml
 type TSOSpec struct {
 	Host                string `yaml:"host"`
@@ -66,7 +69,7 @@ func (s *TSOSpec) Status(ctx context.Context, timeout time.Duration, tlsCfg *tls
 		return "Down"
 	}
 
-	primary, err := pc.GetServicePrimary("tso")
+	primary, err := pc.GetServicePrimary(tsoService)
 	if err != nil {
 		return "ERR"
 	}
@@ -309,6 +312,27 @@ func (i *TSOInstance) setTLSConfig(ctx context.Context, enableTLS bool, configs
 	return configs, nil
 }
 
+// IsPrimary checks if the instance is primary
+func (i *TSOInstance) IsPrimary(ctx context.Context, topo Topology, tlsCfg *tls.Config) (bool, error) {
+	tidbTopo, ok := topo.(*Specification)
+	if !ok {
+		panic("topo should be type of tidb topology")
+	}
+	pdClient := api.NewPDClient(ctx, tidbTopo.GetPDListWithManageHost(), time.Second*5, tlsCfg)
+	primary, err := pdClient.GetServicePrimary(tsoService)
+	if err != nil {
+		return false, errors.Annotatef(err, "failed to get TSO primary %s", i.GetHost())
+	}
+
+	spec := i.InstanceSpec.(*TSOSpec)
+	enableTLS := false
+	if tlsCfg != nil {
+		enableTLS = true
+	}
+
+	return primary == spec.GetAdvertiseListenURL(enableTLS), nil
+}
+
 // ScaleConfig deploy temporary config on scaling
 func (i *TSOInstance) ScaleConfig(
 	ctx context.Context,