enhancement(sysadvisor): export region reclaimable info as non-reclai…

…m when dedicated_cores can not be co-located with reclaimed_cores

Signed-off-by: linzhecheng <linzhecheng@bytedance.com>

pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/region/region_base.go

@@ -106,6 +106,9 @@ type QoSRegionBase struct {
 	// If enableBorweinModel is set, borweinController will update target indicators by model inference.
 	enableBorweinModel bool
 	borweinController  *borweinctrl.BorweinController
+
+	// enableReclaim returns true if the resources of region can be reclaimed to supply for reclaimed_cores
+	enableReclaim func() bool
 }
 
 // NewQoSRegionBase returns a base qos region instance with common region methods
@@ -144,6 +147,7 @@ func NewQoSRegionBase(name string, ownerPoolName string, regionType types.QoSReg
 	if r.enableBorweinModel {
 		r.borweinController = borweinctrl.NewBorweinController(name, regionType, ownerPoolName, conf, metaReader)
 	}
+	r.enableReclaim = r.EnableReclaim
 
 	klog.Infof("[qosaware-cpu] created region [%v/%v/%v]", r.Name(), r.Type(), r.OwnerPoolName())
 
@@ -318,7 +322,7 @@ func (r *QoSRegionBase) GetProvisionPolicy() (policyTopPriority types.CPUProvisi
 		policyTopPriority = r.provisionPolicies[0].name
 	}
 
-	if !r.EnableReclaim {
+	if !r.enableReclaim() {
 		policyInUse = types.CPUProvisionPolicyNonReclaim
 	} else {
 		policyInUse = r.provisionPolicyNameInUse
@@ -336,7 +340,7 @@ func (r *QoSRegionBase) GetHeadRoomPolicy() (policyTopPriority types.CPUHeadroom
 		policyTopPriority = r.headroomPolicies[0].name
 	}
 
-	if !r.EnableReclaim {
+	if !r.enableReclaim() {
 		policyInUse = types.CPUHeadroomPolicyNonReclaim
 	} else {
 		policyInUse = r.headroomPolicyNameInUse
@@ -566,3 +570,7 @@ func (r *QoSRegionBase) updateStatus() {
 		}
 	}
 }
+
+func (r *QoSRegionBase) EnableReclaim() bool {
+	return r.ResourceEssentials.EnableReclaim
+}

pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/region/region_dedicated_numa_exclusive.go

@@ -17,18 +17,23 @@ limitations under the License.
 package region
 
 import (
+	"context"
+	"fmt"
+
 	"k8s.io/apimachinery/pkg/util/uuid"
 	"k8s.io/klog/v2"
 	"k8s.io/kubelet/pkg/apis/resourceplugin/v1alpha1"
 
 	workloadapis "github.com/kubewharf/katalyst-api/pkg/apis/workload/v1alpha1"
 	"github.com/kubewharf/katalyst-core/pkg/agent/qrm-plugins/cpu/dynamicpolicy/state"
 	"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/metacache"
+	"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/plugin/qosaware/resource/helper"
 	"github.com/kubewharf/katalyst-core/pkg/agent/sysadvisor/types"
 	"github.com/kubewharf/katalyst-core/pkg/config"
 	"github.com/kubewharf/katalyst-core/pkg/consts"
 	"github.com/kubewharf/katalyst-core/pkg/metaserver"
 	"github.com/kubewharf/katalyst-core/pkg/metrics"
+	"github.com/kubewharf/katalyst-core/pkg/util/general"
 	"github.com/kubewharf/katalyst-core/pkg/util/machine"
 )
 
@@ -57,10 +62,36 @@ func NewQoSRegionDedicatedNumaExclusive(ci *types.ContainerInfo, conf *config.Co
 	r.indicatorCurrentGetters = map[string]types.IndicatorCurrentGetter{
 		string(workloadapis.TargetIndicatorNameCPI): r.getPodCPICurrent,
 	}
+	r.enableReclaim = r.EnableReclaim
 
 	return r
 }
 
+func (r *QoSRegionDedicatedNumaExclusive) getPodUID() (string, error) {
+	if len(r.podSet) != 1 {
+		return "", fmt.Errorf("more than one pod are assgined to this policy")
+	}
+	for podUID := range r.podSet {
+		return podUID, nil
+	}
+	return "", fmt.Errorf("should never get here")
+}
+
+func (r *QoSRegionDedicatedNumaExclusive) EnableReclaim() bool {
+	podUID, err := r.getPodUID()
+	if err != nil {
+		general.ErrorS(err, "getPodUID failed")
+		return false
+	}
+
+	enableReclaim, err := helper.PodEnableReclaim(context.Background(), r.metaServer, podUID, r.ResourceEssentials.EnableReclaim)
+	if err != nil {
+		general.ErrorS(err, "failed to check PodEnableReclaim", "name", r.name)
+		return false
+	}
+	return enableReclaim
+}
+
 func (r *QoSRegionDedicatedNumaExclusive) TryUpdateProvision() {
 	r.Lock()
 	defer r.Unlock()