github/kubernetes
1
0
Fork 0
mirror of https://github.com/k3s-io/kubernetes.git synced 2025-08-01 15:58:37 +00:00
Code Issues Projects Releases Wiki Activity

horizontal pod autoscaler tolerance breaching verifier

Browse Source
This commit is contained in:
jay vyas 2016-05-06 17:52:30 -04:00
parent 4dc0b8780c
commit db3d1b6942
1 changed files with 69 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

69
pkg/controller/podautoscaler/horizontal_test.go
View File

@ -20,6 +20,7 @@ import (
"encoding/json"
"fmt"
"io"
"math"
"sync"
"testing"
"time"
@ -746,4 +747,72 @@ func TestEventNotCreated(t *testing.T) {
tc.runTest(t)
}
// TestComputedToleranceAlgImplementation is a regression test which
// back-calculates a minimal percentage for downscaling based on a small percentage
// increase in pod utilization which is calibrated against the tolerance value.
func TestComputedToleranceAlgImplementation(t *testing.T) {
startPods := int32(10)
// 150 mCPU per pod.
totalUsedCPUOfAllPods := uint64(startPods * 150)
// Each pod starts out asking for 2X what is really needed.
// This means we will have a 50% ratio of used/requested
totalRequestedCPUOfAllPods := int32(2 * totalUsedCPUOfAllPods)
requestedToUsed := float64(totalRequestedCPUOfAllPods / int32(totalUsedCPUOfAllPods))
// Spread the amount we ask over 10 pods. We can add some jitter later in reportedLevels.
perPodRequested := totalRequestedCPUOfAllPods / startPods
// Force a minimal scaling event by satisfying (tolerance < 1 - resourcesUsedRatio).
target := math.Abs(1/(requestedToUsed*(1-tolerance))) + .01
finalCpuPercentTarget := int32(target * 100)
resourcesUsedRatio := float64(totalUsedCPUOfAllPods) / float64(float64(totalRequestedCPUOfAllPods)*target)
// i.e. .60 * 20 -> scaled down expectation.
finalPods := int32(math.Ceil(resourcesUsedRatio * float64(startPods)))
// To breach tolerance we will create a utilization ratio difference of tolerance to usageRatioToleranceValue)
tc := testCase{
minReplicas: 0,
maxReplicas: 1000,
initialReplicas: startPods,
desiredReplicas: finalPods,
CPUTarget: finalCpuPercentTarget,
reportedLevels: []uint64{
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
totalUsedCPUOfAllPods / 10,
},
reportedCPURequests: []resource.Quantity{
resource.MustParse(fmt.Sprint(perPodRequested+100) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested-100) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested+10) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested-10) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested+2) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested-2) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested+1) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested-1) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested) + "m"),
resource.MustParse(fmt.Sprint(perPodRequested) + "m"),
},
}
tc.runTest(t)
// Reuse the data structure above, now testing "unscaling".
// Now, we test that no scaling happens if we are in a very close margin to the tolerance
target = math.Abs(1/(requestedToUsed*(1-tolerance))) + .004
finalCpuPercentTarget = int32(target * 100)
tc.CPUTarget = finalCpuPercentTarget
tc.initialReplicas = startPods
tc.desiredReplicas = startPods
tc.runTest(t)
}
// TODO: add more tests