fix(sysadvisor): fix failed to find regions

Signed-off-by: linzhecheng <linzhecheng@bytedance.com>

pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/advisor.go

@@ -494,15 +494,39 @@ func (cra *cpuResourceAdvisor) gc() {
 	}
 }
 
-func (cra *cpuResourceAdvisor) getContainerRegions(ci *types.ContainerInfo) ([]region.QoSRegion, error) {
+func (cra *cpuResourceAdvisor) getRegionsByRegionNames(names sets.String) []region.QoSRegion {
 	var regions []region.QoSRegion = nil
-	for regionName := range ci.RegionNames {
-		r, ok := cra.regionMap[regionName]
+	for regionName := range names {
+		region, ok := cra.regionMap[regionName]
 		if !ok {
-			return nil, fmt.Errorf("failed to find region %v", regionName)
+			return nil
 		}
-		regions = append(regions, r)
+		regions = append(regions, region)
 	}
+	return regions
+}
+
+func (cra *cpuResourceAdvisor) getRegionsByPodUID(podUID string) []region.QoSRegion {
+	var regions []region.QoSRegion = nil
+	for _, r := range cra.regionMap {
+		podSet := r.GetPods()
+		for uid := range podSet {
+			if uid == podUID {
+				regions = append(regions, r)
+			}
+		}
+	}
+	return regions
+}
+
+func (cra *cpuResourceAdvisor) getContainerRegions(ci *types.ContainerInfo) ([]region.QoSRegion, error) {
+	regions := cra.getRegionsByRegionNames(ci.RegionNames)
+	if len(regions) > 0 {
+		return regions, nil
+	}
+
+	// The containers of the same pod belong to the same region
+	regions = cra.getRegionsByPodUID(ci.PodUID)
 	return regions, nil
 }
 

pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/region/provisionpolicy/policy_canonical.go

@@ -77,13 +77,14 @@ func (p *PolicyCanonical) Update() error {
 	if err != nil {
 		return err
 	}
-	klog.Infof("[qosaware-cpu-provision] cpu requirement estimation: %.2f, #container %v", cpuEstimation, containerCnt)
 
 	// we need to call SetLatestCPURequirement to ensure the previous requirements are passed to
 	// regulator in case that sysadvisor restarts, to avoid the slow-start always begin with zero.
 	p.regulator.SetLatestCPURequirement(p.requirement)
 	p.regulator.Regulate(cpuEstimation)
 	p.requirement = p.regulator.GetCPURequirement()
+
+	klog.Infof("[qosaware-cpu-provision] cpu requirement estimation: %.2f, requirement: %v #container %v", cpuEstimation, p.requirement, containerCnt)
 	return nil
 }
 

pkg/agent/sysadvisor/plugin/qosaware/resource/cpu/region/regulator/regulator.go

@@ -72,11 +72,11 @@ func (c *CPURegulator) SetLatestCPURequirement(latestCPURequirement int) {
 func (c *CPURegulator) Regulate(cpuRequirementRaw float64) {
 	cpuRequirement := cpuRequirementRaw + float64(c.ReservedForAllocate)
 	cpuRequirement = c.slowdown(cpuRequirement)
-	cpuRequirementInt := c.round(cpuRequirement)
-	cpuRequirementInt = c.clamp(cpuRequirementInt)
+	cpuRequirementRound := c.round(cpuRequirement)
+	cpuRequirementInt := c.clamp(cpuRequirementRound)
 
-	klog.Infof("[qosaware-cpu] cpu requirement by policy: %.2f, after post process: %v, added reserved: %v",
-		cpuRequirementRaw, cpuRequirementInt, c.ReservedForAllocate)
+	klog.Infof("[qosaware-cpu] cpu requirement by policy: %.2f, after slowdown %.2f, after round %v, after post process: %v, added reserved: %v",
+		cpuRequirementRaw, cpuRequirement, cpuRequirementRound, cpuRequirementInt, c.ReservedForAllocate)
 
 	if cpuRequirementInt != c.latestCPURequirement {
 		c.latestCPURequirement = cpuRequirementInt
@@ -96,6 +96,7 @@ func (c *CPURegulator) slowdown(cpuRequirement float64) float64 {
 	// Restrict ramp down period
 	if cpuRequirement < latestCPURequirement && now.Before(c.latestRampDownTime.Add(c.minRampDownPeriod)) {
 		cpuRequirement = latestCPURequirement
+		return cpuRequirement
 	}
 
 	// Restrict ramp up and down step

pkg/agent/sysadvisor/plugin/qosaware/server/cpu/cpu_server.go

@@ -385,13 +385,14 @@ func (cs *cpuServer) removePod(podUID string) error {
 }
 
 func (cs *cpuServer) updatePoolInfo(poolName string, info *cpuadvisor.AllocationInfo) error {
-	// No need to check existence of pool because cpu plugin may return non-exist
-	// pool such as fallback. GC is needed for pool maintenance.
-	pi := &types.PoolInfo{
-		PoolName:                         info.OwnerPoolName,
-		TopologyAwareAssignments:         machine.TransformCPUAssignmentFormat(info.TopologyAwareAssignments),
-		OriginalTopologyAwareAssignments: machine.TransformCPUAssignmentFormat(info.OriginalTopologyAwareAssignments),
+	pi, ok := cs.metaCache.GetPoolInfo(poolName)
+	if !ok {
+		pi = &types.PoolInfo{
+			PoolName: info.OwnerPoolName,
+		}
 	}
+	pi.TopologyAwareAssignments = machine.TransformCPUAssignmentFormat(info.TopologyAwareAssignments)
+	pi.OriginalTopologyAwareAssignments = machine.TransformCPUAssignmentFormat(info.OriginalTopologyAwareAssignments)
 
 	return cs.metaCache.SetPoolInfo(poolName, pi)
 }