Merge pull request #786 from luomingmeng/dev/fix-sysadvisor-reclaim-s…

…tate fix(*): fix sysadvisor state reclaim pod's TopologyAwareAssignments i…
kubewharf · Feb 24, 2025 · 8168777 · 8168777
2 parents 80d36ed + e1f2148
commit 8168777
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 12 deletions.
diff --git a/pkg/agent/qrm-plugins/cpu/dynamicpolicy/policy_advisor_handler.go b/pkg/agent/qrm-plugins/cpu/dynamicpolicy/policy_advisor_handler.go
@@ -158,9 +158,9 @@ func (p *DynamicPolicy) GetCheckpoint(_ context.Context,
 
 			chkEntries[uid].Entries[entryName].OwnerPoolName = ownerPoolName
 
-			// not set topology-aware assignments for shared_cores and reclaimed_cores,
+			// not set topology-aware assignments for shared_cores,
 			// since their topology-aware assignments are same to the pools they are in.
-			if (!allocationInfo.CheckShared() && !allocationInfo.CheckReclaimed()) || containerEntries.IsPoolEntry() {
+			if !allocationInfo.CheckShared() || containerEntries.IsPoolEntry() {
 				chkEntries[uid].Entries[entryName].TopologyAwareAssignments = machine.ParseCPUAssignmentFormat(allocationInfo.TopologyAwareAssignments)
 				chkEntries[uid].Entries[entryName].OriginalTopologyAwareAssignments = machine.ParseCPUAssignmentFormat(allocationInfo.OriginalTopologyAwareAssignments)
 			}
@@ -274,9 +274,9 @@ func (p *DynamicPolicy) createGetAdviceRequest() (*advisorapi.GetAdviceRequest,
 
 			info.AllocationInfo.OwnerPoolName = ownerPoolName
 
-			// not set topology-aware assignments for shared_cores and reclaimed_cores,
+			// not set topology-aware assignments for shared_cores,
 			// since their topology-aware assignments are same to the pools they are in.
-			if (!allocationInfo.CheckShared() && !allocationInfo.CheckReclaimed()) || containerEntries.IsPoolEntry() {
+			if !allocationInfo.CheckShared() || containerEntries.IsPoolEntry() {
 				info.AllocationInfo.TopologyAwareAssignments = machine.ParseCPUAssignmentFormat(allocationInfo.TopologyAwareAssignments)
 				info.AllocationInfo.OriginalTopologyAwareAssignments = machine.ParseCPUAssignmentFormat(allocationInfo.OriginalTopologyAwareAssignments)
 			}

diff --git a/pkg/agent/sysadvisor/plugin/qosaware/resource/memory/headroompolicy/policy_numa_aware.go b/pkg/agent/sysadvisor/plugin/qosaware/resource/memory/headroompolicy/policy_numa_aware.go
@@ -151,28 +151,33 @@ func (p *PolicyNUMAAware) Update() (err error) {
 	if reclaimableMemory > 0 {
 		reduceRatio = p.memoryHeadroom / reclaimableMemory
 	}
-	numaHeadroom := 0.0
+
+	totalNUMAHeadroom := 0.0
+	allNUMAs := p.metaServer.CPUDetails.NUMANodes().ToSliceInt()
+	numaHeadroom := make(map[int]resource.Quantity, len(allNUMAs))
 	for numaID := range numaReclaimableMemory {
 		numaReclaimableMemory[numaID] *= reduceRatio
-		numaHeadroom += numaReclaimableMemory[numaID]
-		p.numaMemoryHeadroom[numaID] = *resource.NewQuantity(int64(numaReclaimableMemory[numaID]), resource.BinarySI)
+		totalNUMAHeadroom += numaReclaimableMemory[numaID]
+		numaHeadroom[numaID] = *resource.NewQuantity(int64(numaReclaimableMemory[numaID]), resource.BinarySI)
 		general.InfoS("memory reclaimable per NUMA", "NUMA-ID", numaID, "headroom", numaReclaimableMemory[numaID])
 	}
 
-	allNUMAs := p.metaServer.CPUDetails.NUMANodes()
-	for _, numaID := range allNUMAs.ToSliceInt() {
-		if _, ok := p.numaMemoryHeadroom[numaID]; !ok {
+	for _, numaID := range allNUMAs {
+		if _, ok := numaHeadroom[numaID]; !ok {
 			general.InfoS("set non-reclaim NUMA memory reclaimable as empty", "NUMA-ID", numaID)
-			p.numaMemoryHeadroom[numaID] = *resource.NewQuantity(0, resource.BinarySI)
+			numaHeadroom[numaID] = *resource.NewQuantity(0, resource.BinarySI)
 		}
 	}
 
+	p.numaMemoryHeadroom = numaHeadroom
+
 	general.InfoS("total memory reclaimable",
 		"reclaimableMemory", general.FormatMemoryQuantity(reclaimableMemory),
 		"ResourceUpperBound", general.FormatMemoryQuantity(p.essentials.ResourceUpperBound),
 		"systemWatermarkReserved", general.FormatMemoryQuantity(systemWatermarkReserved),
 		"reservedForAllocate", general.FormatMemoryQuantity(reservedForAllocate),
 		"headroom", p.memoryHeadroom,
+		"totalNUMAHeadroom", totalNUMAHeadroom,
 		"numaHeadroom", numaHeadroom,
 	)
 	return nil

diff --git a/pkg/agent/sysadvisor/plugin/qosaware/server/cpu_server.go b/pkg/agent/sysadvisor/plugin/qosaware/server/cpu_server.go
@@ -552,7 +552,7 @@ func (cs *cpuServer) setContainerInfoBasedOnAllocationInfo(
 	}
 
 	// fill in topology aware assignment for containers with owner pool
-	if ci.QoSLevel != consts.PodAnnotationQoSLevelDedicatedCores {
+	if info.TopologyAwareAssignments == nil {
 		if len(ci.OwnerPoolName) > 0 {
 			if poolInfo, ok := cs.metaCache.GetPoolInfo(ci.OwnerPoolName); ok {
 				ci.TopologyAwareAssignments = poolInfo.TopologyAwareAssignments.Clone()