Merge pull request #119963 from pohly/dra-scheduler-perf-multiple-claims

dra: scheduler_perf test case with multiple claims per pod

test/integration/scheduler_perf/config/dra/another-resourceclaimtemplate.yaml

@@ -0,0 +1,7 @@
+apiVersion: resource.k8s.io/v1alpha1
+kind: ResourceClaimTemplate
+metadata:
+  name: another-test-claim-template
+spec:
+  spec:
+    resourceClassName: another-test-class

test/integration/scheduler_perf/config/dra/another-resourceclass.yaml

@@ -0,0 +1,5 @@
+apiVersion: resource.k8s.io/v1alpha1
+kind: ResourceClass
+metadata:
+  name: another-test-class
+driverName: another-test-driver.cdi.k8s.io

test/integration/scheduler_perf/config/dra/pod-with-many-claim-templates.yaml

@@ -0,0 +1,27 @@
+apiVersion: v1
+kind: Pod
+metadata:
+  generateName: test-dra
+spec:
+  containers:
+  - image: registry.k8s.io/pause:3.9
+    name: pause
+    resources:
+      claims:
+      - name: resource-1a
+      - name: resource-1b
+      - name: resource-2a
+      - name: resource-2b
+  resourceClaims:
+  - name: resource-1a
+    source:
+      resourceClaimTemplateName: test-claim-template
+  - name: resource-1b
+    source:
+      resourceClaimTemplateName: test-claim-template
+  - name: resource-2a
+    source:
+      resourceClaimTemplateName: another-test-claim-template
+  - name: resource-2b
+    source:
+      resourceClaimTemplateName: another-test-claim-template

test/integration/scheduler_perf/config/dra/resourclass.yaml

@@ -1,5 +0,0 @@
-apiVersion: resource.k8s.io/v1alpha1
-kind: ResourceClass
-metadata:
-  name: scheduler-performance
-driverName: test-driver.cdi.k8s.io

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

@@ -777,3 +777,79 @@
       initPods: 1000
       measurePods: 1000
       maxClaimsPerNode: 10
+
+# This similar to SchedulingWithResourceClaimTemplate, except
+# that it uses four claims per pod, from two different drivers.
+# This emphasizes a bit more the complexity of collaborative
+# scheduling via PodSchedulingContext.
+- name: SchedulingWithMultipleResourceClaims
+  featureGates:
+    DynamicResourceAllocation: true
+  workloadTemplate:
+  - opcode: createNodes
+    countParam: $nodesWithoutDRA
+  - opcode: createNodes
+    nodeTemplatePath: config/dra/node-with-dra-test-driver.yaml
+    countParam: $nodesWithDRA
+  - opcode: createResourceDriver
+    driverName: test-driver.cdi.k8s.io
+    nodes: scheduler-perf-dra-*
+    maxClaimsPerNodeParam: $maxClaimsPerNode
+  - opcode: createResourceDriver
+    driverName: another-test-driver.cdi.k8s.io
+    nodes: scheduler-perf-dra-*
+    maxClaimsPerNodeParam: $maxClaimsPerNode
+  - opcode: createResourceClass
+    templatePath: config/dra/resourceclass.yaml
+  - opcode: createResourceClass
+    templatePath: config/dra/another-resourceclass.yaml
+  - opcode: createResourceClaimTemplate
+    templatePath: config/dra/resourceclaimtemplate.yaml
+    namespace: init
+  - opcode: createResourceClaimTemplate
+    templatePath: config/dra/another-resourceclaimtemplate.yaml
+    namespace: init
+  - opcode: createPods
+    namespace: init
+    countParam: $initPods
+    podTemplatePath: config/dra/pod-with-many-claim-templates.yaml
+  - opcode: createResourceClaimTemplate
+    templatePath: config/dra/resourceclaimtemplate.yaml
+    namespace: test
+  - opcode: createResourceClaimTemplate
+    templatePath: config/dra/another-resourceclaimtemplate.yaml
+    namespace: test
+  - opcode: createPods
+    namespace: test
+    countParam: $measurePods
+    podTemplatePath: config/dra/pod-with-many-claim-templates.yaml
+    collectMetrics: true
+  workloads:
+  - name: fast
+    params:
+      # This testcase runs through all code paths without
+      # taking too long overall.
+      nodesWithDRA: 1
+      nodesWithoutDRA: 1
+      initPods: 0
+      measurePods: 1
+      maxClaimsPerNode: 20
+  - name: 2000pods_100nodes
+    params:
+      # In this testcase, the number of nodes is smaller
+      # than the limit for the PodScheduling slices.
+      nodesWithDRA: 100
+      nodesWithoutDRA: 0
+      initPods: 1000
+      measurePods: 1000
+      maxClaimsPerNode: 40
+  - name: 2000pods_200nodes
+    params:
+      # In this testcase, the driver and scheduler must
+      # truncate the PotentialNodes and UnsuitableNodes
+      # slices.
+      nodesWithDRA: 200
+      nodesWithoutDRA: 0
+      initPods: 1000
+      measurePods: 1000
+      maxClaimsPerNode: 20

test/integration/scheduler_perf/dra_test.go

@@ -210,7 +210,7 @@ func (op *createResourceDriverOp) run(ctx context.Context, tb testing.TB, client
 		}
 	}
 
-	controller := draapp.NewController(clientset, "test-driver.cdi.k8s.io", resources)
+	controller := draapp.NewController(clientset, op.DriverName, resources)
 	ctx, cancel := context.WithCancel(ctx)
 	var wg sync.WaitGroup
 	wg.Add(1)