Define resource tracking mode based on cluster size

Evalle · Nov 12, 2020 · a700174 · a700174
1 parent 3a0d9a9
commit a700174
Show file tree

Hide file tree

Showing 7 changed files with 11 additions and 37 deletions.
diff --git a/clusterloader2/pkg/measurement/common/resource_usage.go b/clusterloader2/pkg/measurement/common/resource_usage.go
@@ -32,6 +32,9 @@ import (
 
 const (
 	resourceUsageMetricName = "ResourceUsageSummary"
+	// maxNodeCountForAllNodes defines the threshold for cluster size above which
+	// we no longer gather resource usage from all system components on all nodes.
+	maxNodeCountForAllNodes = 1000
 )
 
 func init() {
@@ -67,10 +70,6 @@ func (e *resourceUsageMetricMeasurement) Execute(config *measurement.Config) ([]
 		if err != nil {
 			return nil, err
 		}
-		nodeMode, err := util.GetStringOrDefault(config.Params, "nodeMode", "")
-		if err != nil {
-			return nil, err
-		}
 		namespace, err := util.GetStringOrDefault(config.Params, "namespace", "kube-system")
 		if err != nil {
 			return nil, err
@@ -94,19 +93,15 @@ func (e *resourceUsageMetricMeasurement) Execute(config *measurement.Config) ([]
 				}
 			}
 		}
-		var nodesSet gatherers.NodesSet
-		switch nodeMode {
-		case "master":
-			nodesSet = gatherers.MasterNodes
-		case "masteranddns":
-			nodesSet = gatherers.MasterAndDNSNodes
-		case "masterandnondaemons":
+
+		// Compute the node based on the cluster size.
+		nodeCount := config.ClusterFramework.GetClusterConfig().Nodes
+		nodesSet := gatherers.AllNodes
+		if nodeCount > maxNodeCountForAllNodes {
 			nodesSet = gatherers.MasterAndNonDaemons
-		default:
-			nodesSet = gatherers.AllNodes
 		}
 
-		klog.V(2).Infof("%s: starting resource usage collecting...", e)
+		klog.V(2).Infof("%s: starting resource usage collecting (mode %#v)...", e, nodesSet)
 		e.gatherer, err = gatherers.NewResourceUsageGatherer(config.ClusterFramework.GetClientSets().GetClient(), host, config.ClusterFramework.GetClusterConfig().KubeletPort,
 			provider, gatherers.ResourceGathererOptions{
 				InKubemark:                        provider.Name() == "kubemark",

diff --git a/clusterloader2/pkg/measurement/util/gatherers/container_resource_gatherer.go b/clusterloader2/pkg/measurement/util/gatherers/container_resource_gatherer.go
@@ -40,12 +40,8 @@ type NodesSet int
 const (
 	// AllNodes - all containers on all nodes
 	AllNodes NodesSet = 0
-	// MasterNodes - all containers on Master nodes only
-	MasterNodes NodesSet = 1
-	// MasterAndDNSNodes - all containers on Master nodes and DNS containers on other nodes
-	MasterAndDNSNodes NodesSet = 2
 	// MasterAndNonDaemons - all containers on Master nodes and non-daemons on other nodes.
-	MasterAndNonDaemons NodesSet = 3
+	MasterAndNonDaemons NodesSet = 1
 )
 
 // ResourceUsageSummary represents summary of resource usage per container.
@@ -126,12 +122,6 @@ func NewResourceUsageGatherer(c clientset.Interface, host string, port int, prov
 
 		nodesToConsider := make(map[string]bool)
 		for _, pod := range pods.Items {
-			if (options.Nodes == MasterNodes) && !masterNodes.Has(pod.Spec.NodeName) {
-				continue
-			}
-			if (options.Nodes == MasterAndDNSNodes) && !masterNodes.Has(pod.Spec.NodeName) && pod.Labels["k8s-app"] != "kube-dns" {
-				continue
-			}
 			if (options.Nodes == MasterAndNonDaemons) && !masterNodes.Has(pod.Spec.NodeName) && isDaemonPod(&pod) {
 				continue
 			}
@@ -141,7 +131,7 @@ func NewResourceUsageGatherer(c clientset.Interface, host string, port int, prov
 			for _, container := range pod.Status.ContainerStatuses {
 				g.containerIDs = append(g.containerIDs, container.Name)
 			}
-			if options.Nodes == MasterAndDNSNodes || options.Nodes == MasterAndNonDaemons {
+			if options.Nodes == MasterAndNonDaemons {
 				nodesToConsider[pod.Spec.NodeName] = true
 			}
 		}

diff --git a/clusterloader2/testing/access-tokens/config.yaml b/clusterloader2/testing/access-tokens/config.yaml
@@ -36,7 +36,6 @@
 {{$qpsPerWorker := DefaultParam .CL2_ACCESS_TOKENS_QPS (MultiplyFloat 2.5 (DivideFloat .Nodes 5000))}}
 
 # TestMetrics measurement variables
-{{$NODE_MODE := DefaultParam .NODE_MODE "allnodes"}}
 {{$ENABLE_SYSTEM_POD_METRICS:= DefaultParam .ENABLE_SYSTEM_POD_METRICS true}}
 {{$ENABLE_RESTART_COUNT_CHECK := DefaultParam .ENABLE_RESTART_COUNT_CHECK false}}
 {{$RESTART_COUNT_THRESHOLD_OVERRIDES:= DefaultParam .RESTART_COUNT_THRESHOLD_OVERRIDES ""}}
@@ -59,7 +58,6 @@ steps:
       Method: TestMetrics
       Params:
         action: start
-        nodeMode: {{$NODE_MODE}}
         systemPodMetricsEnabled: {{$ENABLE_SYSTEM_POD_METRICS}}
         restartCountThresholdOverrides: {{YamlQuote $RESTART_COUNT_THRESHOLD_OVERRIDES 4}}
         enableRestartCountCheck: {{$ENABLE_RESTART_COUNT_CHECK}}
@@ -174,7 +172,6 @@ steps:
       Method: TestMetrics
       Params:
         action: gather
-        nodeMode: {{$NODE_MODE}}
         systemPodMetricsEnabled: {{$ENABLE_SYSTEM_POD_METRICS}}
         restartCountThresholdOverrides: {{YamlQuote $RESTART_COUNT_THRESHOLD_OVERRIDES 4}}
         enableRestartCountCheck: {{$ENABLE_RESTART_COUNT_CHECK}}
diff --git a/clusterloader2/testing/density/config.yaml b/clusterloader2/testing/density/config.yaml
@@ -4,7 +4,6 @@
 
 #Constants
 {{$DENSITY_RESOURCE_CONSTRAINTS_FILE := DefaultParam .DENSITY_RESOURCE_CONSTRAINTS_FILE ""}}
-{{$NODE_MODE := DefaultParam .NODE_MODE "allnodes"}}
 {{$NODES_PER_NAMESPACE := DefaultParam .NODES_PER_NAMESPACE 100}}
 {{$PODS_PER_NODE := DefaultParam .PODS_PER_NODE 30}}
 {{$DENSITY_TEST_THROUGHPUT := DefaultParam .DENSITY_TEST_THROUGHPUT 20}}
@@ -87,7 +86,6 @@ steps:
     Method: TestMetrics
     Params:
       action: start
-      nodeMode: {{$NODE_MODE}}
       resourceConstraints: {{$DENSITY_RESOURCE_CONSTRAINTS_FILE}}
       systemPodMetricsEnabled: {{$ENABLE_SYSTEM_POD_METRICS}}
       clusterOOMsTrackerEnabled: {{$ENABLE_CLUSTER_OOMS_TRACKER}}

diff --git a/clusterloader2/testing/density/high-density-config.yaml b/clusterloader2/testing/density/high-density-config.yaml
@@ -5,7 +5,6 @@
 
 #Constants
 {{$DENSITY_RESOURCE_CONSTRAINTS_FILE := DefaultParam .DENSITY_RESOURCE_CONSTRAINTS_FILE ""}}
-{{$NODE_MODE := DefaultParam .NODE_MODE "allnodes"}}
 {{$NODES_PER_NAMESPACE := DefaultParam .NODES_PER_NAMESPACE 100}}
 {{$PODS_PER_NODE := DefaultParam .PODS_PER_NODE 30}}
 {{$DENSITY_TEST_THROUGHPUT := DefaultParam .DENSITY_TEST_THROUGHPUT 20}}
@@ -79,7 +78,6 @@ steps:
     Method: TestMetrics
     Params:
       action: start
-      nodeMode: {{$NODE_MODE}}
       resourceConstraints: {{$DENSITY_RESOURCE_CONSTRAINTS_FILE}}
       systemPodMetricsEnabled: {{$ENABLE_SYSTEM_POD_METRICS}}
       restartCountThresholdOverrides: {{YamlQuote $RESTART_COUNT_THRESHOLD_OVERRIDES 4}}

diff --git a/clusterloader2/testing/experimental/storage/pod-startup/config.yaml b/clusterloader2/testing/experimental/storage/pod-startup/config.yaml
@@ -3,7 +3,6 @@
 # - If using Persistent Volumes, the default storage class must have volumeBindingMode: Immediate
 
 # Cluster Variables
-  {{$NODE_MODE := DefaultParam .NODE_MODE "allnodes"}}
   {{$NODES_PER_NAMESPACE := DefaultParam .NODES_PER_NAMESPACE 100}}
 
 # Test Variales
@@ -51,7 +50,6 @@ steps:
     Method: TestMetrics
     Params:
       action: start
-      nodeMode: {{$NODE_MODE}}
   - Identifier: PodWithVolumesStartupLatency
     Method: PodStartupLatency
     Params:

diff --git a/clusterloader2/testing/load/config.yaml b/clusterloader2/testing/load/config.yaml
@@ -5,7 +5,6 @@
 # - Only half of Deployments will be assigned 1-1 to existing SVCs.
 
 #Constants
-{{$NODE_MODE := DefaultParam .NODE_MODE "allnodes"}}
 {{$NODES_PER_NAMESPACE := DefaultParam .NODES_PER_NAMESPACE 100}}
 {{$PODS_PER_NODE := DefaultParam .PODS_PER_NODE 30}}
 {{$LOAD_TEST_THROUGHPUT := DefaultParam .CL2_LOAD_TEST_THROUGHPUT 10}}
@@ -153,7 +152,6 @@ steps:
     Method: TestMetrics
     Params:
       action: start
-      nodeMode: {{$NODE_MODE}}
       systemPodMetricsEnabled: {{$ENABLE_SYSTEM_POD_METRICS}}
       clusterOOMsTrackerEnabled: {{$ENABLE_CLUSTER_OOMS_TRACKER}}
       clusterOOMsIgnoredProcesses: {{$CLUSTER_OOMS_IGNORED_PROCESSES}}