scheduler: make device topology alignment switchable (#2044)

Signed-off-by: wangjianyu.wjy <wangjianyu.wjy@alibaba-inc.com> Co-authored-by: wangjianyu.wjy <wangjianyu.wjy@alibaba-inc.com>
koordinator-sh · May 25, 2024 · f1afba1 · f1afba1
1 parent b8892e0
commit f1afba1
Show file tree

Hide file tree

Showing 5 changed files with 38 additions and 12 deletions.
diff --git a/pkg/scheduler/apis/config/types.go b/pkg/scheduler/apis/config/types.go
@@ -219,4 +219,6 @@ type DeviceShareArgs struct {
 	Allocator string
 	// ScoringStrategy selects the device resource scoring strategy.
 	ScoringStrategy *ScoringStrategy
+	// DisableDeviceNUMATopologyAlignment indicates device don't need to align with other resources' numa topology
+	DisableDeviceNUMATopologyAlignment bool
 }
diff --git a/pkg/scheduler/apis/config/v1beta3/types.go b/pkg/scheduler/apis/config/v1beta3/types.go
@@ -214,4 +214,6 @@ type DeviceShareArgs struct {
 	Allocator string `json:"allocator,omitempty"`
 	// ScoringStrategy selects the device resource scoring strategy.
 	ScoringStrategy *ScoringStrategy `json:"scoringStrategy,omitempty"`
+	// DisableDeviceNUMATopologyAlignment indicates device don't need to align with other resources' numa topology
+	DisableDeviceNUMATopologyAlignment bool `json:"disableDeviceNUMATopologyAlignment,omitempty"`
 }
diff --git a/pkg/scheduler/apis/config/v1beta3/zz_generated.conversion.go b/pkg/scheduler/apis/config/v1beta3/zz_generated.conversion.go
diff --git a/pkg/scheduler/plugins/deviceshare/plugin.go b/pkg/scheduler/plugins/deviceshare/plugin.go
@@ -67,9 +67,10 @@ var (
 )
 
 type Plugin struct {
-	handle          frameworkext.ExtendedHandle
-	nodeDeviceCache *nodeDeviceCache
-	scorer          *resourceAllocationScorer
+	disableDeviceNUMATopologyAlignment bool
+	handle                             frameworkext.ExtendedHandle
+	nodeDeviceCache                    *nodeDeviceCache
+	scorer                             *resourceAllocationScorer
 }
 
 type preFilterState struct {
@@ -292,8 +293,11 @@ func (p *Plugin) Filter(ctx context.Context, cycleState *framework.CycleState, p
 	restoreState := reservationRestoreState.getNodeState(node.Name)
 	preemptible := appendAllocated(nil, restoreState.mergedUnmatchedUsed, state.preemptibleDevices[node.Name])
 
-	store := topologymanager.GetStore(cycleState)
-	affinity := store.GetAffinity(node.Name)
+	var affinity topologymanager.NUMATopologyHint
+	if !p.disableDeviceNUMATopologyAlignment {
+		store := topologymanager.GetStore(cycleState)
+		affinity = store.GetAffinity(nodeInfo.Node().Name)
+	}
 
 	allocator := &AutopilotAllocator{
 		state:      state,
@@ -302,6 +306,9 @@ func (p *Plugin) Filter(ctx context.Context, cycleState *framework.CycleState, p
 		pod:        pod,
 		numaNodes:  affinity.NUMANodeAffinity,
 	}
+	if !p.disableDeviceNUMATopologyAlignment {
+		allocator.numaNodes = nil
+	}
 
 	nodeDeviceInfo.lock.RLock()
 	defer nodeDeviceInfo.lock.RUnlock()
@@ -354,8 +361,11 @@ func (p *Plugin) FilterReservation(ctx context.Context, cycleState *framework.Cy
 		return nil
 	}
 
-	store := topologymanager.GetStore(cycleState)
-	affinity := store.GetAffinity(nodeInfo.Node().Name)
+	var affinity topologymanager.NUMATopologyHint
+	if !p.disableDeviceNUMATopologyAlignment {
+		store := topologymanager.GetStore(cycleState)
+		affinity = store.GetAffinity(nodeInfo.Node().Name)
+	}
 
 	allocator := &AutopilotAllocator{
 		state:      state,
@@ -393,8 +403,11 @@ func (p *Plugin) Reserve(ctx context.Context, cycleState *framework.CycleState,
 		return nil
 	}
 
-	store := topologymanager.GetStore(cycleState)
-	affinity := store.GetAffinity(nodeInfo.Node().Name)
+	var affinity topologymanager.NUMATopologyHint
+	if !p.disableDeviceNUMATopologyAlignment {
+		store := topologymanager.GetStore(cycleState)
+		affinity = store.GetAffinity(nodeInfo.Node().Name)
+	}
 
 	allocator := &AutopilotAllocator{
 		state:      state,
@@ -559,8 +572,9 @@ func New(obj runtime.Object, handle framework.Handle) (framework.Plugin, error)
 	go deviceCache.gcNodeDevice(context.TODO(), handle.SharedInformerFactory(), defaultGCPeriod)
 
 	return &Plugin{
-		handle:          extendedHandle,
-		nodeDeviceCache: deviceCache,
-		scorer:          scorePlugin(args),
+		handle:                             extendedHandle,
+		nodeDeviceCache:                    deviceCache,
+		scorer:                             scorePlugin(args),
+		disableDeviceNUMATopologyAlignment: args.DisableDeviceNUMATopologyAlignment,
 	}, nil
 }
diff --git a/pkg/scheduler/plugins/deviceshare/topology_hint.go b/pkg/scheduler/plugins/deviceshare/topology_hint.go
@@ -31,6 +31,9 @@ import (
 )
 
 func (p *Plugin) GetPodTopologyHints(ctx context.Context, cycleState *framework.CycleState, pod *corev1.Pod, nodeName string) (map[string][]topologymanager.NUMATopologyHint, *framework.Status) {
+	if p.disableDeviceNUMATopologyAlignment {
+		return nil, nil
+	}
 	state, status := getPreFilterState(cycleState)
 	if !status.IsSuccess() {
 		return nil, status
@@ -55,6 +58,9 @@ func (p *Plugin) GetPodTopologyHints(ctx context.Context, cycleState *framework.
 }
 
 func (p *Plugin) Allocate(ctx context.Context, cycleState *framework.CycleState, affinity topologymanager.NUMATopologyHint, pod *corev1.Pod, nodeName string) *framework.Status {
+	if p.disableDeviceNUMATopologyAlignment {
+		return nil
+	}
 	state, status := getPreFilterState(cycleState)
 	if !status.IsSuccess() {
 		return status