slo-controller: let node-level network bandwidth takes precedence ove…

…r cluster or node strategy Signed-off-by: sjtufl <jerryfang555@qq.com>
koordinator-sh · Apr 7, 2024 · 1170f25 · 1170f25
1 parent 8838ce9
commit 1170f25
Show file tree

Hide file tree

Showing 2 changed files with 70 additions and 6 deletions.
diff --git a/pkg/slo-controller/nodeslo/resource_strategy.go b/pkg/slo-controller/nodeslo/resource_strategy.go
@@ -25,6 +25,7 @@ import (
 	"k8s.io/klog/v2"
 
 	"github.com/koordinator-sh/koordinator/apis/configuration"
+	"github.com/koordinator-sh/koordinator/apis/extension"
 	slov1alpha1 "github.com/koordinator-sh/koordinator/apis/slo/v1alpha1"
 	"github.com/koordinator-sh/koordinator/pkg/util"
 )
@@ -79,6 +80,9 @@ func getCPUBurstConfigSpec(node *corev1.Node, cfg *configuration.CPUBurstCfg) (*
 }
 
 func getSystemConfigSpec(node *corev1.Node, cfg *configuration.SystemCfg) (*slov1alpha1.SystemStrategy, error) {
+	var nodeSystemConfig *slov1alpha1.SystemStrategy
+
+	// Find strategy that matches current node.
 	nodeLabels := labels.Set(node.Labels)
 	for _, nodeStrategy := range cfg.NodeStrategies {
 		selector, err := metav1.LabelSelectorAsSelector(nodeStrategy.NodeSelector)
@@ -87,11 +91,29 @@ func getSystemConfigSpec(node *corev1.Node, cfg *configuration.SystemCfg) (*slov
 			continue
 		}
 		if selector.Matches(nodeLabels) {
-			return nodeStrategy.SystemStrategy.DeepCopy(), nil
+			nodeSystemConfig = nodeStrategy.SystemStrategy.DeepCopy()
+			break
 		}
+	}
 
+	// If there is no matched node strategy, use cluster strategy.
+	if nodeSystemConfig == nil {
+		nodeSystemConfig = cfg.ClusterStrategy.DeepCopy()
 	}
-	return cfg.ClusterStrategy.DeepCopy(), nil
+
+	// Check whether node bandwidth is specified on current node, which takes HIGHER priority
+	// than ones configured in cluster strategy or node strategy.
+	// Error is returned if failed to parse node total bandwidth from annotation, which is not
+	// supposed to happen because we will check the validity of the annotation value in node
+	// plugins of validating webhook.
+	if nodeBandwidthQuantity, err := extension.GetNodeTotalBandwidth(node.Annotations); err != nil {
+		klog.Errorf("failed to get node total bandwidth from annotation, error: %v", err)
+		return nil, err
+	} else if nodeBandwidthQuantity != nil {
+		nodeSystemConfig.TotalNetworkBandwidth = *nodeBandwidthQuantity
+	}
+
+	return nodeSystemConfig, nil
 }
 
 func getHostApplicationConfig(node *corev1.Node, cfg *configuration.HostApplicationCfg) ([]slov1alpha1.HostApplicationSpec, error) {

diff --git a/pkg/slo-controller/nodeslo/resource_strategy_test.go b/pkg/slo-controller/nodeslo/resource_strategy_test.go
@@ -28,6 +28,7 @@ import (
 	"k8s.io/utils/pointer"
 
 	"github.com/koordinator-sh/koordinator/apis/configuration"
+	"github.com/koordinator-sh/koordinator/apis/extension"
 	ext "github.com/koordinator-sh/koordinator/apis/extension"
 	slov1alpha1 "github.com/koordinator-sh/koordinator/apis/slo/v1alpha1"
 	"github.com/koordinator-sh/koordinator/pkg/util/sloconfig"
@@ -812,12 +813,14 @@ func Test_getSystemConfigSpec(t *testing.T) {
 	defaultConfig := DefaultSLOCfg()
 	testingSystemConfig := &configuration.SystemCfg{
 		ClusterStrategy: &slov1alpha1.SystemStrategy{
-			MinFreeKbytesFactor: pointer.Int64(150),
+			MinFreeKbytesFactor:   pointer.Int64(150),
+			TotalNetworkBandwidth: resource.MustParse("10G"),
 		},
 	}
 	testingSystemConfig1 := &configuration.SystemCfg{
 		ClusterStrategy: &slov1alpha1.SystemStrategy{
-			MinFreeKbytesFactor: pointer.Int64(150),
+			MinFreeKbytesFactor:   pointer.Int64(150),
+			TotalNetworkBandwidth: resource.MustParse("100M"),
 		},
 		NodeStrategies: []configuration.NodeSystemStrategy{
 			{
@@ -829,7 +832,8 @@ func Test_getSystemConfigSpec(t *testing.T) {
 					},
 				},
 				SystemStrategy: &slov1alpha1.SystemStrategy{
-					MinFreeKbytesFactor: pointer.Int64(120),
+					MinFreeKbytesFactor:   pointer.Int64(120),
+					TotalNetworkBandwidth: resource.MustParse("10M"),
 				},
 			},
 			{
@@ -841,11 +845,16 @@ func Test_getSystemConfigSpec(t *testing.T) {
 					},
 				},
 				SystemStrategy: &slov1alpha1.SystemStrategy{
-					MinFreeKbytesFactor: pointer.Int64(130),
+					MinFreeKbytesFactor:   pointer.Int64(130),
+					TotalNetworkBandwidth: resource.MustParse("1000M"),
 				},
 			},
 		},
 	}
+	injectNodeBandwidth := func(systemStrategy *slov1alpha1.SystemStrategy, bandwidth resource.Quantity) *slov1alpha1.SystemStrategy {
+		systemStrategy.TotalNetworkBandwidth = bandwidth
+		return systemStrategy
+	}
 	type args struct {
 		node *corev1.Node
 		cfg  *configuration.SystemCfg
@@ -907,6 +916,39 @@ func Test_getSystemConfigSpec(t *testing.T) {
 			},
 			want: testingSystemConfig1.NodeStrategies[0].SystemStrategy,
 		},
+		{
+			name: "use node-wise bandwidth config while use cluster strategy",
+			args: args{
+				node: &corev1.Node{
+					ObjectMeta: metav1.ObjectMeta{
+						Name: "test-node",
+						Annotations: map[string]string{
+							extension.AnnotationNodeBandwidth: "99M",
+						},
+					},
+				},
+				cfg: testingSystemConfig,
+			},
+			want: injectNodeBandwidth(testingSystemConfig.ClusterStrategy.DeepCopy(), resource.MustParse("99M")),
+		},
+		{
+			name: "use node-wise bandwidth config while use node strategy",
+			args: args{
+				node: &corev1.Node{
+					ObjectMeta: metav1.ObjectMeta{
+						Name: "test-node",
+						Labels: map[string]string{
+							"zzz": "zzz",
+						},
+						Annotations: map[string]string{
+							extension.AnnotationNodeBandwidth: "99M",
+						},
+					},
+				},
+				cfg: testingSystemConfig1,
+			},
+			want: injectNodeBandwidth(testingSystemConfig1.NodeStrategies[1].SystemStrategy.DeepCopy(), resource.MustParse("99M")),
+		},
 	}
 	for _, tt := range tests {
 		t.Run(tt.name, func(t *testing.T) {