Merge branch 'main' into hormes-patch-1

koordinator-sh · Sep 2, 2024 · 0841acf · 0841acf
2 parents 9b2cb8d + cb89d6d
commit 0841acf
Show file tree

Hide file tree

Showing 102 changed files with 5,998 additions and 552 deletions.
diff --git a/.github/workflows/e2e-k8s-1.22.yaml b/.github/workflows/e2e-k8s-1.22.yaml
@@ -23,6 +23,23 @@ jobs:
     continue-on-error: true
     runs-on: ubuntu-20.04
     steps:
+      - name: Check host environment before
+        run: |
+          set -ex
+          lscpu -e
+          tree -L 2 /sys/
+          tree -L 2 /sys/fs/cgroup
+          df -h
+      - name: Free Disk Space
+        uses: jlumbroso/free-disk-space@v1.3.1
+        with:
+          tool-cache: false
+          swap-storage: false
+          large-packages: false
+          docker-images: false
+          android: true
+          dotnet: true
+          haskell: true
       - uses: actions/checkout@v4
         with:
           submodules: true
@@ -48,15 +65,12 @@ jobs:
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${MANAGER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${MANAGER_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${KOORDLET_IMAGE} || { echo >&2 "kind not installed or error loading image: ${KOORDLET_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${SCHEDULER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${SCHEDULER_IMAGE}"; exit 1; }
-      - name: Check host environment
+      - name: Check cluster environment
         run: |
           set -ex
           kubectl version --short
           kubectl get pods -A
           kubectl get nodes -o yaml
-          tree -L 2 /sys/
-          tree -L 2 /sys/fs/cgroup
-          cat /proc/cpuinfo
       - name: Install Koordinator
         run: |
           set -ex

diff --git a/.github/workflows/e2e-k8s-1.24.yaml b/.github/workflows/e2e-k8s-1.24.yaml
@@ -23,6 +23,23 @@ jobs:
     continue-on-error: true
     runs-on: ubuntu-20.04
     steps:
+      - name: Check host environment before
+        run: |
+          set -ex
+          lscpu -e
+          tree -L 2 /sys/
+          tree -L 2 /sys/fs/cgroup
+          df -h
+      - name: Free Disk Space
+        uses: jlumbroso/free-disk-space@v1.3.1
+        with:
+          tool-cache: false
+          swap-storage: false
+          large-packages: false
+          docker-images: false
+          android: true
+          dotnet: true
+          haskell: true
       - uses: actions/checkout@v4
         with:
           submodules: true
@@ -48,15 +65,12 @@ jobs:
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${MANAGER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${MANAGER_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${KOORDLET_IMAGE} || { echo >&2 "kind not installed or error loading image: ${KOORDLET_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${SCHEDULER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${SCHEDULER_IMAGE}"; exit 1; }
-      - name: Check host environment
+      - name: Check cluster environment
         run: |
           set -ex
           kubectl version --short
           kubectl get pods -A
           kubectl get nodes -o yaml
-          tree -L 2 /sys/
-          tree -L 2 /sys/fs/cgroup
-          cat /proc/cpuinfo
       - name: Install Koordinator
         run: |
           set -ex

diff --git a/.github/workflows/e2e-k8s-1.28.yaml b/.github/workflows/e2e-k8s-1.28.yaml
@@ -23,6 +23,23 @@ jobs:
     continue-on-error: true
     runs-on: ubuntu-20.04
     steps:
+      - name: Check host environment before
+        run: |
+          set -ex
+          lscpu -e
+          tree -L 2 /sys/
+          tree -L 2 /sys/fs/cgroup
+          df -h
+      - name: Free Disk Space
+        uses: jlumbroso/free-disk-space@v1.3.1
+        with:
+          tool-cache: false
+          swap-storage: false
+          large-packages: false
+          docker-images: false
+          android: true
+          dotnet: true
+          haskell: true
       - uses: actions/checkout@v4
         with:
           submodules: true
@@ -48,15 +65,12 @@ jobs:
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${MANAGER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${MANAGER_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${KOORDLET_IMAGE} || { echo >&2 "kind not installed or error loading image: ${KOORDLET_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${SCHEDULER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${SCHEDULER_IMAGE}"; exit 1; }
-      - name: Check host environment
+      - name: Check cluster environment
         run: |
           set -ex
           kubectl version --short
           kubectl get pods -A
           kubectl get nodes -o yaml
-          tree -L 2 /sys/
-          tree -L 2 /sys/fs/cgroup
-          cat /proc/cpuinfo
       - name: Install Koordinator
         run: |
           set -ex

diff --git a/.github/workflows/e2e-k8s-latest.yaml b/.github/workflows/e2e-k8s-latest.yaml
@@ -22,6 +22,23 @@ jobs:
     continue-on-error: true
     runs-on: ubuntu-20.04
     steps:
+      - name: Check host environment before
+        run: |
+          set -ex
+          lscpu -e
+          tree -L 2 /sys/
+          tree -L 2 /sys/fs/cgroup
+          df -h
+      - name: Free Disk Space
+        uses: jlumbroso/free-disk-space@v1.3.1
+        with:
+          tool-cache: false
+          swap-storage: false
+          large-packages: false
+          docker-images: false
+          android: true
+          dotnet: true
+          haskell: true
       - uses: actions/checkout@v4
         with:
           submodules: true
@@ -46,15 +63,12 @@ jobs:
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${MANAGER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${MANAGER_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${KOORDLET_IMAGE} || { echo >&2 "kind not installed or error loading image: ${KOORDLET_IMAGE}"; exit 1; }
           kind load docker-image --name=${KIND_CLUSTER_NAME} ${SCHEDULER_IMAGE} || { echo >&2 "kind not installed or error loading image: ${SCHEDULER_IMAGE}"; exit 1; }
-      - name: Check host environment
+      - name: Check cluster environment
         run: |
           set -ex
           kubectl version --short
           kubectl get pods -A
           kubectl get nodes -o yaml
-          tree -L 2 /sys/
-          tree -L 2 /sys/fs/cgroup
-          cat /proc/cpuinfo
       - name: Install Koordinator
         run: |
           set -ex
@@ -105,3 +119,5 @@ jobs:
               exit 1
           fi
           exit $retVal
+      - name: Checkout host environment after
+        run: df -h
diff --git a/.licenseignore b/.licenseignore
@@ -4,6 +4,8 @@ pkg/descheduler/controllers/migration/controllerfinder
 pkg/scheduler/plugins/coscheduling
 pkg/scheduler/plugins/nodenumaresource/least_allocated.go
 pkg/scheduler/plugins/nodenumaresource/most_allocated.go
+pkg/scheduler/plugins/reservation/preemption.go
+pkg/scheduler/plugins/reservation/preemption_test.go
 pkg/scheduler/frameworkext/topologymanager/policy.go
 pkg/scheduler/frameworkext/topologymanager/policy_test.go
 pkg/scheduler/frameworkext/topologymanager/policy_best_effort.go

diff --git a/apis/configuration/slo_controller_config.go b/apis/configuration/slo_controller_config.go
@@ -18,20 +18,22 @@ package configuration
 
 import (
 	"github.com/mohae/deepcopy"
+	corev1 "k8s.io/api/core/v1"
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
 
 	slov1alpha1 "github.com/koordinator-sh/koordinator/apis/slo/v1alpha1"
 )
 
 const (
 	// keys in the configmap
-	ColocationConfigKey        = "colocation-config"
-	ResourceThresholdConfigKey = "resource-threshold-config"
-	ResourceQOSConfigKey       = "resource-qos-config"
-	CPUBurstConfigKey          = "cpu-burst-config"
-	SystemConfigKey            = "system-config"
-	HostApplicationConfigKey   = "host-application-config"
-	CPUNormalizationConfigKey  = "cpu-normalization-config"
+	ColocationConfigKey            = "colocation-config"
+	ResourceThresholdConfigKey     = "resource-threshold-config"
+	ResourceQOSConfigKey           = "resource-qos-config"
+	CPUBurstConfigKey              = "cpu-burst-config"
+	SystemConfigKey                = "system-config"
+	HostApplicationConfigKey       = "host-application-config"
+	CPUNormalizationConfigKey      = "cpu-normalization-config"
+	ResourceAmplificationConfigKey = "resource-amplification-config"
 )
 
 // +k8s:deepcopy-gen=true
@@ -295,6 +297,29 @@ type ModelRatioCfg struct {
 	HyperThreadTurboEnabledRatio *float64 `json:"hyperThreadTurboEnabledRatio,omitempty"`
 }
 
+// ResourceAmplificationCfg is the cluster-level configuration of the resource amplification strategy.
+// +k8s:deepcopy-gen=true
+type ResourceAmplificationCfg struct {
+	ResourceAmplificationStrategy `json:",inline"`
+	NodeConfigs                   []NodeResourceAmplificationCfg `json:"nodeConfigs,omitempty" validate:"dive"`
+}
+
+// NodeResourceAmplificationCfg is the node-level configuration of the resource amplification strategy.
+// +k8s:deepcopy-gen=true
+type NodeResourceAmplificationCfg struct {
+	NodeCfgProfile `json:",inline"`
+	ResourceAmplificationStrategy
+}
+
+// ResourceAmplificationStrategy is the resource amplification strategy.
+// +k8s:deepcopy-gen=true
+type ResourceAmplificationStrategy struct {
+	// Enable defines whether the resource amplification strategy is enabled.
+	Enable *bool `json:"enable,omitempty"`
+	// ResourceAmplificationRatio defines resource amplification ratio
+	ResourceAmplificationRatio map[corev1.ResourceName]float64 `json:"resourceAmplificationRatio,omitempty"`
+}
+
 /*
 Koordinator uses configmap to manage the configuration of SLO, the configmap is stored in
  <ConfigNameSpace>/<SLOCtrlConfigMap>, with the following keys respectively:

diff --git a/apis/configuration/zz_generated.deepcopy.go b/apis/configuration/zz_generated.deepcopy.go
diff --git a/apis/extension/preemption.go b/apis/extension/preemption.go
@@ -23,6 +23,9 @@ const (
 	// PreemptionPolicyTransformer is enabled.
 	// When PreemptionPolicyTransformer is enabled but the pod does not set the label, DefaultPreemptionPolicy is used.
 	LabelPodPreemptionPolicy = SchedulingDomainPrefix + "/preemption-policy"
+
+	// LabelDisablePreemptible determines whether the pod disables being a victim during the reservation preemption.
+	LabelDisablePreemptible = SchedulingDomainPrefix + "/disable-preemptible"
 )
 
 func GetPodKoordPreemptionPolicy(pod *corev1.Pod) *corev1.PreemptionPolicy {
@@ -40,3 +43,14 @@ func GetPodKoordPreemptionPolicy(pod *corev1.Pod) *corev1.PreemptionPolicy {
 func GetPreemptionPolicyPtr(policy corev1.PreemptionPolicy) *corev1.PreemptionPolicy {
 	return &policy
 }
+
+func IsPodPreemptible(pod *corev1.Pod) bool {
+	if pod == nil || pod.Labels == nil {
+		return true
+	}
+	v, ok := pod.Labels[LabelDisablePreemptible]
+	if !ok {
+		return true
+	}
+	return v != "true"
+}