fix(qrm): fix cpu admit failed caused by float ceil

Makefile

@@ -122,7 +122,7 @@ vet: ## Run go vet against code.
 
 .PHONY: test
 test: ## Run go test against code.
-	go test -v -coverprofile=coverage.txt -parallel=16 -p=16 -covermode=atomic -race -coverpkg=./... \
+	go test -v -json -coverprofile=coverage.txt -parallel=16 -p=16 -covermode=atomic -race -coverpkg=./... \
 		`go list ./pkg/... | grep -E -v "pkg/scheduler|pkg/controller/resource-recommend|pkg/util/resource-recommend"`
 
 .PHONY: license

cmd/katalyst-agent/app/options/qrm/memory_plugin.go

@@ -25,14 +25,15 @@ import (
 )
 
 type MemoryOptions struct {
-	PolicyName                  string
-	ReservedMemoryGB            uint64
-	SkipMemoryStateCorruption   bool
-	EnableSettingMemoryMigrate  bool
-	EnableMemoryAdvisor         bool
-	ExtraControlKnobConfigFile  string
-	EnableOOMPriority           bool
-	OOMPriorityPinnedMapAbsPath string
+	PolicyName                                    string
+	ReservedMemoryGB                              uint64
+	SkipMemoryStateCorruption                     bool
+	EnableSettingMemoryMigrate                    bool
+	EnableMemoryAdvisor                           bool
+	ExtraControlKnobConfigFile                    string
+	EnableOOMPriority                             bool
+	OOMPriorityPinnedMapAbsPath                   string
+	EnableNonBindingShareCoresMemoryResourceCheck bool
 
 	SockMemOptions
 	LogCacheOptions
@@ -71,6 +72,7 @@ func NewMemoryOptions() *MemoryOptions {
 		EnableSettingMemoryMigrate: false,
 		EnableMemoryAdvisor:        false,
 		EnableOOMPriority:          false,
+		EnableNonBindingShareCoresMemoryResourceCheck: true,
 		SockMemOptions: SockMemOptions{
 			EnableSettingSockMem: false,
 			SetGlobalTCPMemRatio: 20,  // default: 20% * {host total memory}
@@ -105,6 +107,8 @@ func (o *MemoryOptions) AddFlags(fss *cliflag.NamedFlagSets) {
 		o.ExtraControlKnobConfigFile, "the absolute path of extra control knob config file")
 	fs.BoolVar(&o.EnableOOMPriority, "enable-oom-priority",
 		o.EnableOOMPriority, "if set true, we will enable oom priority enhancement")
+	fs.BoolVar(&o.EnableNonBindingShareCoresMemoryResourceCheck, "enable-non-binding-share-cores-memory-resource-check",
+		o.EnableNonBindingShareCoresMemoryResourceCheck, "enable the topology check for non-binding shares cores pods")
 	fs.StringVar(&o.OOMPriorityPinnedMapAbsPath, "oom-priority-pinned-bpf-map-path",
 		o.OOMPriorityPinnedMapAbsPath, "the absolute path of oom priority pinned bpf map")
 	fs.BoolVar(&o.EnableSettingSockMem, "enable-setting-sockmem",
@@ -137,6 +141,7 @@ func (o *MemoryOptions) ApplyTo(conf *qrmconfig.MemoryQRMPluginConfig) error {
 	conf.EnableMemoryAdvisor = o.EnableMemoryAdvisor
 	conf.ExtraControlKnobConfigFile = o.ExtraControlKnobConfigFile
 	conf.EnableOOMPriority = o.EnableOOMPriority
+	conf.EnableNonBindingShareCoresMemoryResourceCheck = o.EnableNonBindingShareCoresMemoryResourceCheck
 	conf.OOMPriorityPinnedMapAbsPath = o.OOMPriorityPinnedMapAbsPath
 	conf.EnableSettingSockMem = o.EnableSettingSockMem
 	conf.SetGlobalTCPMemRatio = o.SetGlobalTCPMemRatio

pkg/agent/qrm-plugins/cpu/dynamicpolicy/policy.go

@@ -1173,7 +1173,7 @@ func (p *DynamicPolicy) getContainerRequestedCores(allocationInfo *state.Allocat
 	return allocationInfo.RequestQuantity
 }
 
-func (p *DynamicPolicy) checkNormalShareCoresCpuResource(req *pluginapi.ResourceRequest) (bool, error) {
+func (p *DynamicPolicy) checkNonBindingShareCoresCpuResource(req *pluginapi.ResourceRequest) (bool, error) {
 	_, reqFloat64, err := util.GetPodAggregatedRequestResource(req)
 	if err != nil {
 		return false, fmt.Errorf("GetQuantityFromResourceReq failed with error: %v", err)
@@ -1186,14 +1186,14 @@ func (p *DynamicPolicy) checkNormalShareCoresCpuResource(req *pluginapi.Resource
 		state.WrapAllocationMetaFilter((*commonstate.AllocationMeta).CheckDedicated),
 		state.WrapAllocationMetaFilter((*commonstate.AllocationMeta).CheckSharedOrDedicatedNUMABinding))
 
-	general.Infof("[checkNormalShareCoresCpuResource] node cpu allocated: %d, allocatable: %d", shareCoresAllocatedInt, pooledCPUs.Size())
+	general.Infof("[checkNonBindingShareCoresCpuResource] node cpu allocated: %d, allocatable: %d", shareCoresAllocatedInt, pooledCPUs.Size())
 	if shareCoresAllocatedInt > pooledCPUs.Size() {
-		general.Warningf("[checkNormalShareCoresCpuResource] no enough cpu resource for normal share cores pod: %s/%s, container: %s (request: %.02f, node allocated: %d, node allocatable: %d)",
+		general.Warningf("[checkNonBindingShareCoresCpuResource] no enough cpu resource for non-binding share cores pod: %s/%s, container: %s (request: %.02f, node allocated: %d, node allocatable: %d)",
 			req.PodNamespace, req.PodName, req.ContainerName, reqFloat64, shareCoresAllocatedInt, pooledCPUs.Size())
 		return false, nil
 	}
 
-	general.InfoS("checkNormalShareCoresCpuResource memory successfully",
+	general.InfoS("checkNonBindingShareCoresCpuResource memory successfully",
 		"podNamespace", req.PodNamespace,
 		"podName", req.PodName,
 		"containerName", req.ContainerName,

pkg/agent/qrm-plugins/cpu/dynamicpolicy/policy_advisor_handler.go

@@ -656,7 +656,7 @@ func (p *DynamicPolicy) applyBlocks(blockCPUSet advisorapi.BlockCPUSet, resp *ad
 					if allocationInfo.CheckSharedNUMABinding() {
 						poolEntry.QoSLevel = apiconsts.PodAnnotationQoSLevelSharedCores
 						// set SharedNUMABinding declarations to pool entry containing SharedNUMABinding containers,
-						// in order to differentiate them from normal share pools during GetFilteredPoolsCPUSetMap.
+						// in order to differentiate them from non-binding share cores pools during GetFilteredPoolsCPUSetMap.
 						poolEntry.Annotations = general.MergeMap(poolEntry.Annotations, map[string]string{
 							apiconsts.PodAnnotationMemoryEnhancementNumaBinding: apiconsts.PodAnnotationMemoryEnhancementNumaBindingEnable,
 						})

pkg/agent/qrm-plugins/cpu/dynamicpolicy/policy_allocation_handlers.go

@@ -767,7 +767,7 @@ func (p *DynamicPolicy) calcPoolResizeRequest(originAllocation, allocation *stat
 			allocation.PodNamespace, allocation.PodName, allocation.ContainerName, originPodAggregatedRequest, podAggregatedRequest)
 	}
 
-	// only support normal share and snb inplace update resize now
+	// only support share cores inplace update resize now (include non-binding share cores and share cores with NUMA binding)
 	if allocation.CheckSharedNUMABinding() {
 		// check snb numa migrate for inplace update resize
 		originTargetNumaID, err := state.GetSharedNUMABindingTargetNuma(originAllocation)
@@ -1136,7 +1136,7 @@ func (p *DynamicPolicy) applyPoolsAndIsolatedInfo(poolsCPUSet map[string]machine
 					if allocationInfo.CheckSharedNUMABinding() {
 						poolEntry.QoSLevel = apiconsts.PodAnnotationQoSLevelSharedCores
 						// set SharedNUMABinding declarations to pool entry containing SharedNUMABinding containers,
-						// in order to differentiate them from normal share pools during GetFilteredPoolsCPUSetMap.
+						// in order to differentiate them from non-binding share cores pools during GetFilteredPoolsCPUSetMap.
 						poolEntry.Annotations = general.MergeMap(poolEntry.Annotations, map[string]string{
 							apiconsts.PodAnnotationMemoryEnhancementNumaBinding: apiconsts.PodAnnotationMemoryEnhancementNumaBindingEnable,
 						})