kubernetes
/
autoscaler
mirror of https://github.com/kubernetes/autoscaler.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Support gpus in nodes and pods definitions in UT

This commit is contained in:
Łukasz Osipiuk 2018-05-11 18:00:23 +02:00 committed by Łukasz Osipiuk
parent be381facfb
commit c406da4174
4 changed files with 105 additions and 56 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

54
cluster-autoscaler/core/scale_down_test.go
View File

@ -869,10 +869,10 @@ var defaultScaleDownOptions = context.AutoscalingOptions{
func TestScaleDownEmptyMultipleNodeGroups(t *testing.T) {
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1_1", 1000, 1000, true, "ng1"},
{"n1_2", 1000, 1000, true, "ng1"},
{"n2_1", 1000, 1000, true, "ng2"},
{"n2_2", 1000, 1000, true, "ng2"},
{"n1_1", 1000, 1000, 0, true, "ng1"},
{"n1_2", 1000, 1000, 0, true, "ng1"},
{"n2_1", 1000, 1000, 0, true, "ng2"},
{"n2_2", 1000, 1000, 0, true, "ng2"},
},
options: defaultScaleDownOptions,
expectedScaleDowns: []string{"n1_1", "n2_1"},
@ -883,8 +883,8 @@ func TestScaleDownEmptyMultipleNodeGroups(t *testing.T) {
func TestScaleDownEmptySingleNodeGroup(t *testing.T) {
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 1000, 1000, true, "ng1"},
{"n2", 1000, 1000, true, "ng1"},
{"n1", 1000, 1000, 0, true, "ng1"},
{"n2", 1000, 1000, 0, true, "ng1"},
},
options: defaultScaleDownOptions,
expectedScaleDowns: []string{"n1"},
@ -897,8 +897,8 @@ func TestScaleDownEmptyMinCoresLimitHit(t *testing.T) {
options.MinCoresTotal = 2
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 2000, 1000, true, "ng1"},
{"n2", 1000, 1000, true, "ng1"},
{"n1", 2000, 1000, 0, true, "ng1"},
{"n2", 1000, 1000, 0, true, "ng1"},
},
options: options,
expectedScaleDowns: []string{"n2"},
@ -911,10 +911,10 @@ func TestScaleDownEmptyMinMemoryLimitHit(t *testing.T) {
options.MinMemoryTotal = 4000
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 2000, 1000 * MB, true, "ng1"},
{"n2", 1000, 1000 * MB, true, "ng1"},
{"n3", 1000, 1000 * MB, true, "ng1"},
{"n4", 1000, 3000 * MB, true, "ng1"},
{"n1", 2000, 1000 * MB, 0, true, "ng1"},
{"n2", 1000, 1000 * MB, 0, true, "ng1"},
{"n3", 1000, 1000 * MB, 0, true, "ng1"},
{"n4", 1000, 3000 * MB, 0, true, "ng1"},
},
options: options,
expectedScaleDowns: []string{"n1", "n2"},
@ -926,7 +926,7 @@ func TestScaleDownEmptyMinGroupSizeLimitHit(t *testing.T) {
options := defaultScaleDownOptions
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 2000, 1000, true, "ng1"},
{"n1", 2000, 1000, 0, true, "ng1"},
},
options: options,
expectedScaleDowns: []string{},
@ -1279,13 +1279,13 @@ func TestCleanUpNodeAutoprovisionedGroups(t *testing.T) {
func TestCalculateCoresAndMemoryTotal(t *testing.T) {
nodeConfigs := []nodeConfig{
{"n1", 2000, 7500 * MB, true, "ng1"},
{"n2", 2000, 7500 * MB, true, "ng1"},
{"n3", 2000, 7500 * MB, true, "ng1"},
{"n4", 12000, 8000 * MB, true, "ng1"},
{"n5", 16000, 7500 * MB, true, "ng1"},
{"n6", 8000, 6000 * MB, true, "ng1"},
{"n7", 6000, 16000 * MB, true, "ng1"},
{"n1", 2000, 7500 * MB, 0, true, "ng1"},
{"n2", 2000, 7500 * MB, 0, true, "ng1"},
{"n3", 2000, 7500 * MB, 0, true, "ng1"},
{"n4", 12000, 8000 * MB, 0, true, "ng1"},
{"n5", 16000, 7500 * MB, 0, true, "ng1"},
{"n6", 8000, 6000 * MB, 0, true, "ng1"},
{"n7", 6000, 16000 * MB, 0, true, "ng1"},
}
nodes := make([]*apiv1.Node, len(nodeConfigs))
for i, n := range nodeConfigs {
@ -1310,13 +1310,13 @@ func TestCalculateCoresAndMemoryTotal(t *testing.T) {
func TestFilterOutMasters(t *testing.T) {
nodeConfigs := []nodeConfig{
{"n1", 2000, 4000, false, "ng1"},
{"n2", 2000, 4000, true, "ng2"},
{"n3", 2000, 8000, true, ""}, // real master
{"n4", 1000, 2000, true, "ng3"},
{"n5", 1000, 2000, true, "ng3"},
{"n6", 2000, 8000, true, ""}, // same machine type, no node group, no api server
{"n7", 2000, 8000, true, ""}, // real master
{"n1", 2000, 4000, 0, false, "ng1"},
{"n2", 2000, 4000, 0, true, "ng2"},
{"n3", 2000, 8000, 0, true, ""}, // real master
{"n4", 1000, 2000, 0, true, "ng3"},
{"n5", 1000, 2000, 0, true, "ng3"},
{"n6", 2000, 8000, 0, true, ""}, // same machine type, no node group, no api server
{"n7", 2000, 8000, 0, true, ""}, // real master
}
nodes := make([]*apiv1.Node, len(nodeConfigs))
for i, n := range nodeConfigs {

2
cluster-autoscaler/core/scale_test_common.go
View File

@ -24,6 +24,7 @@ type nodeConfig struct {
name string
cpu int64
memory int64
gpu int64
ready bool
group string
}
@ -32,6 +33,7 @@ type podConfig struct {
name string
cpu int64
memory int64
gpu int64
node string
}

71
cluster-autoscaler/core/scale_up_test.go
View File

@ -58,15 +58,15 @@ var defaultOptions = context.AutoscalingOptions{
func TestScaleUpOK(t *testing.T) {
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 100, 100, true, "ng1"},
{"n2", 1000, 1000, true, "ng2"},
{"n1", 100, 100, 0, true, "ng1"},
{"n2", 1000, 1000, 0, true, "ng2"},
},
pods: []podConfig{
{"p1", 80, 0, "n1"},
{"p2", 800, 0, "n2"},
{"p1", 80, 0, 0, "n1"},
{"p2", 800, 0, 0, "n2"},
},
extraPods: []podConfig{
{"p-new", 500, 0, ""},
{"p-new", 500, 0, 0, ""},
},
scaleUpOptionToChoose: groupSizeChange{groupName: "ng2", sizeChange: 1},
expectedFinalScaleUp: groupSizeChange{groupName: "ng2", sizeChange: 1},
@ -81,16 +81,16 @@ func TestScaleUpMaxCoresLimitHit(t *testing.T) {
options.MaxCoresTotal = 9
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 2000, 100, true, "ng1"},
{"n2", 4000, 1000, true, "ng2"},
{"n1", 2000, 100, 0, true, "ng1"},
{"n2", 4000, 1000, 0, true, "ng2"},
},
pods: []podConfig{
{"p1", 1000, 0, "n1"},
{"p2", 3000, 0, "n2"},
{"p1", 1000, 0, 0, "n1"},
{"p2", 3000, 0, 0, "n2"},
},
extraPods: []podConfig{
{"p-new-1", 2000, 0, ""},
{"p-new-2", 2000, 0, ""},
{"p-new-1", 2000, 0, 0, ""},
{"p-new-2", 2000, 0, 0, ""},
},
scaleUpOptionToChoose: groupSizeChange{groupName: "ng1", sizeChange: 2},
expectedFinalScaleUp: groupSizeChange{groupName: "ng1", sizeChange: 1},
@ -107,17 +107,17 @@ func TestScaleUpMaxMemoryLimitHit(t *testing.T) {
options.MaxMemoryTotal = 1300 // set in mb
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 2000, 100 * MB, true, "ng1"},
{"n2", 4000, 1000 * MB, true, "ng2"},
{"n1", 2000, 100 * MB, 0, true, "ng1"},
{"n2", 4000, 1000 * MB, 0, true, "ng2"},
},
pods: []podConfig{
{"p1", 1000, 0, "n1"},
{"p2", 3000, 0, "n2"},
{"p1", 1000, 0, 0, "n1"},
{"p2", 3000, 0, 0, "n2"},
},
extraPods: []podConfig{
{"p-new-1", 2000, 100 * MB, ""},
{"p-new-2", 2000, 100 * MB, ""},
{"p-new-3", 2000, 100 * MB, ""},
{"p-new-1", 2000, 100 * MB, 0, ""},
{"p-new-2", 2000, 100 * MB, 0, ""},
{"p-new-3", 2000, 100 * MB, 0, ""},
},
scaleUpOptionToChoose: groupSizeChange{groupName: "ng1", sizeChange: 3},
expectedFinalScaleUp: groupSizeChange{groupName: "ng1", sizeChange: 2},
@ -132,17 +132,17 @@ func TestScaleUpCapToMaxTotalNodesLimit(t *testing.T) {
options.MaxNodesTotal = 3
config := &scaleTestConfig{
nodes: []nodeConfig{
{"n1", 2000, 100 * MB, true, "ng1"},
{"n2", 4000, 1000 * MB, true, "ng2"},
{"n1", 2000, 100 * MB, 0, true, "ng1"},
{"n2", 4000, 1000 * MB, 0, true, "ng2"},
},
pods: []podConfig{
{"p1", 1000, 0, "n1"},
{"p2", 3000, 0, "n2"},
{"p1", 1000, 0, 0, "n1"},
{"p2", 3000, 0, 0, "n2"},
},
extraPods: []podConfig{
{"p-new-1", 4000, 100 * MB, ""},
{"p-new-2", 4000, 100 * MB, ""},
{"p-new-3", 4000, 100 * MB, ""},
{"p-new-1", 4000, 100 * MB, 0, ""},
{"p-new-2", 4000, 100 * MB, 0, ""},
{"p-new-3", 4000, 100 * MB, 0, ""},
},
scaleUpOptionToChoose: groupSizeChange{groupName: "ng2", sizeChange: 3},
expectedFinalScaleUp: groupSizeChange{groupName: "ng2", sizeChange: 1},
@ -211,6 +211,9 @@ func simpleScaleUpTest(t *testing.T, config *scaleTestConfig) {
nodes := make([]*apiv1.Node, len(config.nodes))
for i, n := range config.nodes {
node := BuildTestNode(n.name, n.cpu, n.memory)
if n.gpu > 0 {
AddGpusToNode(node, n.gpu)
}
SetNodeReadyState(node, n.ready, time.Now())
nodes[i] = node
groups[n.group] = append(groups[n.group], node)
@ -218,9 +221,8 @@ func simpleScaleUpTest(t *testing.T, config *scaleTestConfig) {
pods := make(map[string][]apiv1.Pod)
for _, p := range config.pods {
pod := *BuildTestPod(p.name, p.cpu, p.memory)
pod.Spec.NodeName = p.node
pods[p.node] = append(pods[p.node], pod)
pod := buildTestPod(p)
pods[p.node] = append(pods[p.node], *pod)
}
fakeClient.Fake.AddReactor("list", "pods", func(action core.Action) (bool, runtime.Object, error) {
@ -276,7 +278,7 @@ func simpleScaleUpTest(t *testing.T, config *scaleTestConfig) {
extraPods := make([]*apiv1.Pod, len(config.extraPods))
for i, p := range config.extraPods {
pod := BuildTestPod(p.name, p.cpu, p.memory)
pod := buildTestPod(p)
extraPods[i] = pod
}
@ -312,6 +314,17 @@ func getGroupSizeChangeFromChan(c chan groupSizeChange) *groupSizeChange {
}
}
func buildTestPod(p podConfig) *apiv1.Pod {
pod := BuildTestPod(p.name, p.cpu, p.memory)
if p.gpu > 0 {
RequestGpuForPod(pod, p.gpu)
}
if p.node != "" {
pod.Spec.NodeName = p.node
}
return pod
}
func TestScaleUpNodeComingNoScale(t *testing.T) {
n1 := BuildTestNode("n1", 100, 1000)
SetNodeReadyState(n1, true, time.Now())

34
cluster-autoscaler/utils/test/test_utils.go
View File

@ -64,6 +64,26 @@ func BuildTestPod(name string, cpu int64, mem int64) *apiv1.Pod {
return pod
}
const (
// cannot use constants from gpu module due to cyclic package import
resourceNvidiaGPU = "nvidia.com/gpu"
gpuLabel = "cloud.google.com/gke-accelerator"
defaultGPUType = "nvidia-tesla-k80"
)
// RequestGpuForPod modifies pod's resource requests by adding a number of GPUs to them.
func RequestGpuForPod(pod *apiv1.Pod, gpusCount int64) {
if pod.Spec.Containers[0].Resources.Limits == nil {
pod.Spec.Containers[0].Resources.Limits = apiv1.ResourceList{}
}
pod.Spec.Containers[0].Resources.Limits[resourceNvidiaGPU] = *resource.NewQuantity(gpusCount, resource.DecimalSI)
if pod.Spec.Containers[0].Resources.Requests == nil {
pod.Spec.Containers[0].Resources.Requests = apiv1.ResourceList{}
}
pod.Spec.Containers[0].Resources.Requests[resourceNvidiaGPU] = *resource.NewQuantity(gpusCount, resource.DecimalSI)
}
// BuildTestNode creates a node with specified capacity.
func BuildTestNode(name string, millicpu int64, mem int64) *apiv1.Node {
node := &apiv1.Node{
@ -97,6 +117,20 @@ func BuildTestNode(name string, millicpu int64, mem int64) *apiv1.Node {
return node
}
// AddGpusToNode adds GPU capacity to given node. Default accelerator type is used.
func AddGpusToNode(node *apiv1.Node, gpusCount int64) {
node.Spec.Taints = append(
node.Spec.Taints,
apiv1.Taint{
Key: resourceNvidiaGPU,
Value: "present",
Effect: "NoSchedule",
})
node.Status.Capacity[resourceNvidiaGPU] = *resource.NewQuantity(gpusCount, resource.DecimalSI)
node.Status.Allocatable[resourceNvidiaGPU] = *resource.NewQuantity(gpusCount, resource.DecimalSI)
node.Labels[gpuLabel] = defaultGPUType
}
// SetNodeReadyState sets node ready state.
func SetNodeReadyState(node *apiv1.Node, ready bool, lastTransition time.Time) {
for i := range node.Status.Conditions {