Merge pull request #127825 from macsko/add_pod_update_event_handling_scheduler_perf_test_case

Add scheduler_perf test case for pod update events handling

test/integration/scheduler_perf/config/event_handling/podupdate-node-unschedulable.yaml

@@ -0,0 +1,15 @@
+apiVersion: v1
+kind: Node
+metadata:
+  name: unschedulable-node
+spec:
+  unschedulable: true
+status:
+  capacity:
+    pods: "90000"
+    cpu: "4"
+    memory: 32Gi
+  conditions:
+    - status: "True"
+      type: Ready
+  phase: Running

test/integration/scheduler_perf/config/event_handling/podupdate-node-with-taint.yaml

@@ -0,0 +1,17 @@
+apiVersion: v1
+kind: Node
+metadata:
+  name: node-with-taint
+spec:
+  taints:
+    - key: toleration
+      effect: NoSchedule
+status:
+  capacity:
+    pods: "90000"
+    cpu: "4"
+    memory: 32Gi
+  conditions:
+    - status: "True"
+      type: Ready
+  phase: Running

test/integration/scheduler_perf/config/event_handling/podupdate-pod-blocker-update.yaml

@@ -0,0 +1,13 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-blocker-{{ .Index }}
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause
+    resources:
+      requests:
+        cpu: 0.0001
+        memory: 1Mi
+  nodeName: scheduler-perf-node

test/integration/scheduler_perf/config/event_handling/podupdate-pod-blocker.yaml

@@ -0,0 +1,15 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-blocker-{{ .Index }}
+  labels:
+    color: green-{{ .Index }}
+    topology: blue
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause
+    resources:
+      requests:
+        cpu: 0.0001
+        memory: {{ div 30000 .Count }}Mi

test/integration/scheduler_perf/config/event_handling/podupdate-pod-interpodaffinity.yaml

@@ -0,0 +1,19 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  generateName: pod-interpodaffinity-
+  labels:
+    color: green-{{ .Index }}
+    type: unsched
+spec:
+  affinity:
+    podAntiAffinity:
+      requiredDuringSchedulingIgnoredDuringExecution:
+      - labelSelector:
+          matchLabels:
+            color: green-{{ .Index }}
+        topologyKey: kubernetes.io/hostname
+        namespaces: [blocker]
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-noderesources.yaml

@@ -0,0 +1,14 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-noderesources-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause
+    resources:
+      requests:
+        cpu: 0.0001
+        memory: {{ div 30000 .Count }}Mi

test/integration/scheduler_perf/config/event_handling/podupdate-pod-nodeunschedulable-update.yaml

@@ -0,0 +1,14 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-nodeunschedulable-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  tolerations:
+  - key: node.kubernetes.io/unschedulable
+    operator: Exists
+    effect: NoSchedule
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-nodeunschedulable.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-nodeunschedulable-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-podtopologyspread.yaml

@@ -0,0 +1,19 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  generateName: pod-podtopologyspread-
+  labels:
+    type: unsched
+    topology: blue
+spec:
+  topologySpreadConstraints:
+    - maxSkew: {{ .Count }}
+      minDomains: 100
+      topologyKey: kubernetes.io/hostname
+      whenUnsatisfiable: DoNotSchedule
+      labelSelector:
+        matchLabels:
+          topology: blue
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-schedulinggates-update.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-schedulinggates-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-schedulinggates.yaml

@@ -0,0 +1,12 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-schedulinggates-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  schedulingGates:
+  - name: test.k8s.io/hold
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-tainttoleration-update.yaml

@@ -0,0 +1,14 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-tainttoleration-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  tolerations:
+  - key: toleration
+    operator: Exists
+    effect: NoSchedule
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/event_handling/podupdate-pod-tainttoleration.yaml

@@ -0,0 +1,10 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  name: pod-tainttoleration-{{ .Index }}
+  labels:
+    type: unsched
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.10
+    name: pause

test/integration/scheduler_perf/config/performance-config.yaml

@@ -1663,3 +1663,124 @@
       initNodes: 50
       blockerPods: 480 # Must be slightly below initNodes * 10 to be stable
       measurePods: 500 # Must be initNodes * 10
+
+# This test case is used to measure the performance of queuing hints when handling the pod update events: 
+# UpdatePodLabel, UpdatePodScaleDown, UpdatePodTolerations and UpdatePodSchedulingGatesEliminated.
+# It has a few stages, but general idea is to make a node and block some pods on it
+# or to create additional blocker pods that will prevent the other ones from being scheduled.
+# Then, updating the blocker pods or the unschedulable pods themselves generate cluster events,
+# that through QHints make the pods schedulable. 
+# Plugins covered: InterPodAffinity, NodeResources, NodeUnschedulable, PodTopologySpread, SchedulingGates and TaintToleration.
+- name: EventHandlingPodUpdate
+  featureGates:
+    SchedulerQueueingHints: true
+    InPlacePodVerticalScaling: true
+  workloadTemplate:
+  # Collect metrics from all createPods ops that initially create unschedulable pods (type: unsched).
+  - opcode: startCollectingMetrics
+    namespaces: [nodeunschedulable, tainttoleration, blocker, interpodaffinity, noderesources, schedulinggates]
+    labelSelector:
+      type: unsched
+  # Create one unschedulable node.
+  - opcode: createNodes
+    count: 1
+    nodeTemplatePath: config/event_handling/podupdate-node-unschedulable.yaml
+  # Created pods blocked using NodeUnschedulable plugin.
+  - opcode: createPods
+    countParam: $measurePods
+    podTemplatePath: config/event_handling/podupdate-pod-nodeunschedulable.yaml
+    skipWaitToCompletion: true
+    namespace: nodeunschedulable
+  # Create one node with NoSchedule taint.
+  - opcode: createNodes
+    count: 1
+    nodeTemplatePath: config/event_handling/podupdate-node-with-taint.yaml
+  # Created pods blocked using TaintToleration plugin.
+  - opcode: createPods
+    countParam: $measurePods
+    podTemplatePath: config/event_handling/podupdate-pod-tainttoleration.yaml
+    skipWaitToCompletion: true
+    namespace: tainttoleration
+  # Wait for unschedulable pods to be processed by the scheduler.
+  - opcode: barrier
+    stageRequirement: Attempted
+  # Update pods blocked using NodeUnschedulable plugin to make them schedulable.
+  - opcode: updateAny
+    countParam: $measurePods
+    templatePath: config/event_handling/podupdate-pod-nodeunschedulable-update.yaml
+    updatePerSecond: 100
+    namespace: nodeunschedulable
+  # Update pods blocked using TaintToleration plugin to make them schedulable.
+  - opcode: updateAny
+    countParam: $measurePods
+    templatePath: config/event_handling/podupdate-pod-tainttoleration-update.yaml
+    updatePerSecond: 100
+    namespace: tainttoleration
+  # Wait for NodeUnschedulable and TaintToleration pods to be scheduled.
+  - opcode: barrier
+  # Create schedulable nodes.
+  - opcode: createNodes
+    count: 1
+    nodeTemplatePath: config/templates/node-with-name.yaml
+  # Create pods that will block other pods from being scheduled.
+  # They'll block using InterPodAffinity, NodeResources and PodTopologySpread plugins.
+  # All blocker pods are scheduled before proceeding.
+  - opcode: createPods
+    countParam: $blockerPods
+    podTemplatePath: config/event_handling/podupdate-pod-blocker.yaml
+    namespace: blocker
+  # Created pods blocked using InterPodAffinity plugin.
+  - opcode: createPods
+    countParam: $measurePods
+    podTemplatePath: config/event_handling/podupdate-pod-interpodaffinity.yaml
+    skipWaitToCompletion: true
+    namespace: interpodaffinity
+  # Created pods blocked using NodeResources plugin.
+  - opcode: createPods
+    countParam: $measurePods
+    podTemplatePath: config/event_handling/podupdate-pod-noderesources.yaml
+    skipWaitToCompletion: true
+    namespace: noderesources
+  # Created pods blocked using PodTopologySpread plugin.
+  # Count has to match $blockerPods as pod uses it as a maxSkew value, 
+  # that must be equal to number of blocker pods.
+  - opcode: createPods
+    countParam: $blockerPods
+    podTemplatePath: config/event_handling/podupdate-pod-podtopologyspread.yaml
+    skipWaitToCompletion: true
+    namespace: blocker
+  # Created pods blocked using SchedulingGates plugin.
+  - opcode: createPods
+    countParam: $measurePods
+    podTemplatePath: config/event_handling/podupdate-pod-schedulinggates.yaml
+    skipWaitToCompletion: true
+    namespace: schedulinggates
+  # Wait for unschedulable pods to be processed by the scheduler.
+  - opcode: barrier
+    stageRequirement: Attempted
+    labelSelector:
+      type: unsched
+  # Update blocker pods' labels and scale down their resource requests 
+  # to make the unschedulable pods schedulable.
+  - opcode: updateAny
+    countParam: $blockerPods
+    templatePath: config/event_handling/podupdate-pod-blocker-update.yaml
+    updatePerSecond: 100
+    namespace: blocker
+  # Update pods blocked by SchedulingGates by removing the gate from themselves.
+  - opcode: updateAny
+    countParam: $measurePods
+    templatePath: config/event_handling/podupdate-pod-schedulinggates-update.yaml
+    updatePerSecond: 100
+    namespace: schedulinggates
+  # Wait for previously unschedulable pods to be scheduled.
+  - opcode: barrier
+    labelSelector:
+      type: unsched
+  - opcode: stopCollectingMetrics
+  workloads:
+  - name: 1Node_1000Pods
+    labels: [performance, short]
+    params:
+      blockerPods: 1000
+      measurePods: 1000