github/kubernetes
4
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #20368 from mqliang/cleanup-unhealthy-replicas

Browse Source 
Auto commit by PR queue bot
This commit is contained in:
k8s-merge-robot
2016-02-08 20:18:25 -08:00
parent 242000d790 86aea1d59c
commit a2c8712aee
2 changed files with 405 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

168
pkg/controller/deployment/deployment_controller.go
View File

@@ -813,20 +813,11 @@ func (dc *DeploymentController) reconcileNewRC(allRCs []*api.ReplicationControll
func (dc *DeploymentController) reconcileOldRCs(allRCs []*api.ReplicationController, oldRCs []*api.ReplicationController, newRC *api.ReplicationController, deployment extensions.Deployment, expectationsCheck bool) (bool, error) {
oldPodsCount := deploymentutil.GetReplicaCountForRCs(oldRCs)
if oldPodsCount == 0 {
// Cant scale down further
// Can't scale down further
return false, nil
}
maxUnavailable, isPercent, err := util.GetIntOrPercentValue(&deployment.Spec.Strategy.RollingUpdate.MaxUnavailable)
if err != nil {
return false, fmt.Errorf("invalid value for MaxUnavailable: %v", err)
}
if isPercent {
maxUnavailable = util.GetValueFromPercent(maxUnavailable, deployment.Spec.Replicas)
}
// Check if we can scale down.
minAvailable := deployment.Spec.Replicas - maxUnavailable
minReadySeconds := deployment.Spec.MinReadySeconds
// Check the expectations of deployment before counting available pods
// Check the expectations of deployment before reconciling
dKey, err := controller.KeyFunc(&deployment)
if err != nil {
return false, fmt.Errorf("Couldn't get key for deployment %#v: %v", deployment, err)
@@ -835,20 +826,144 @@ func (dc *DeploymentController) reconcileOldRCs(allRCs []*api.ReplicationControl
glog.V(4).Infof("Pod expectations not met yet before reconciling old RCs\n")
return false, nil
}
// Find the number of ready pods.
readyPodCount, err := deploymentutil.GetAvailablePodsForRCs(dc.client, allRCs, minReadySeconds)
minReadySeconds := deployment.Spec.MinReadySeconds
allPodsCount := deploymentutil.GetReplicaCountForRCs(allRCs)
newRCAvailablePodCount, err := deploymentutil.GetAvailablePodsForRCs(dc.client, []*api.ReplicationController{newRC}, minReadySeconds)
if err != nil {
return false, fmt.Errorf("could not find available pods: %v", err)
}
if readyPodCount <= minAvailable {
// Cannot scale down.
maxUnavailable, isPercent, err := util.GetIntOrPercentValue(&deployment.Spec.Strategy.RollingUpdate.MaxUnavailable)
if err != nil {
return false, fmt.Errorf("invalid value for MaxUnavailable: %v", err)
}
if isPercent {
maxUnavailable = util.GetValueFromPercent(maxUnavailable, deployment.Spec.Replicas)
}
// Check if we can scale down. We can scale down in the following 2 cases:
// * Some old rcs have unhealthy replicas, we could safely scale down those unhealthy replicas since that won't further
// increase unavailability.
// * New rc has scaled up and it's replicas becomes ready, then we can scale down old rcs in a further step.
//
// maxScaledDown := allPodsCount - minAvailable - newRCPodsUnavailable
// take into account not only maxUnavailable and any surge pods that have been created, but also unavailable pods from
// the newRC, so that the unavailable pods from the newRC would not make us scale down old RCs in a further step(that will
// increase unavailability).
//
// Concrete example:
//
// * 10 replicas
// * 2 maxUnavailable (absolute number, not percent)
// * 3 maxSurge (absolute number, not percent)
//
// case 1:
// * Deployment is updated, newRC is created with 3 replicas, oldRC is scaled down to 8, and newRC is scaled up to 5.
// * The new RC pods crashloop and never become available.
// * allPodsCount is 13. minAvailable is 8. newRCPodsUnavailable is 5.
// * A node fails and causes one of the oldRC pods to become unavailable. However, 13 - 8 - 5 = 0, so the oldRC won't be scaled down.
// * The user notices the crashloop and does kubectl rollout undo to rollback.
// * newRCPodsUnavailable is 1, since we rolled back to the good RC, so maxScaledDown = 13 - 8 - 1 = 4. 4 of the crashlooping pods will be scaled down.
// * The total number of pods will then be 9 and the newRC can be scaled up to 10.
//
// case 2:
// Same example, but pushing a new pod template instead of rolling back (aka "roll over"):
// * The new RC created must start with 0 replicas because allPodsCount is already at 13.
// * However, newRCPodsUnavailable would also be 0, so the 2 old RCs could be scaled down by 5 (13 - 8 - 0), which would then
// allow the new RC to be scaled up by 5.
minAvailable := deployment.Spec.Replicas - maxUnavailable
newRCUnavailablePodCount := newRC.Spec.Replicas - newRCAvailablePodCount
maxScaledDown := allPodsCount - minAvailable - newRCUnavailablePodCount
if maxScaledDown <= 0 {
return false, nil
}
totalScaleDownCount := readyPodCount - minAvailable
// Clean up unhealthy replicas first, otherwise unhealthy replicas will block deployment
// and cause timeout. See https://github.com/kubernetes/kubernetes/issues/16737
cleanupCount, err := dc.cleanupUnhealthyReplicas(oldRCs, deployment, maxScaledDown)
if err != nil {
return false, nil
}
// Scale down old rcs, need check maxUnavailable to ensure we can scale down
scaledDownCount, err := dc.scaleDownOldRCsForRollingUpdate(allRCs, oldRCs, deployment)
if err != nil {
return false, nil
}
totalScaledDown := cleanupCount + scaledDownCount
if expectationsCheck {
dc.podExpectations.ExpectDeletions(dKey, totalScaledDown)
}
return totalScaledDown > 0, nil
}
// cleanupUnhealthyReplicas will scale down old rcs with unhealthy replicas, so that all unhealthy replicas will be deleted.
func (dc *DeploymentController) cleanupUnhealthyReplicas(oldRCs []*api.ReplicationController, deployment extensions.Deployment, maxCleanupCount int) (int, error) {
sort.Sort(controller.ControllersByCreationTimestamp(oldRCs))
// Safely scale down all old rcs with unhealthy replicas. ReplicationController/ReplicaSet will sort the pods in the order
// such that not-ready < ready, unscheduled < scheduled, and pending < running. This ensures that unhealthy replicas will
// been deleted first and won't increase unavailability.
totalScaledDown := 0
for _, targetRC := range oldRCs {
if totalScaleDownCount == 0 {
if totalScaledDown >= maxCleanupCount {
break
}
if targetRC.Spec.Replicas == 0 {
// cannot scale down this RC.
continue
}
readyPodCount, err := deploymentutil.GetAvailablePodsForRCs(dc.client, []*api.ReplicationController{targetRC}, 0)
if err != nil {
return totalScaledDown, fmt.Errorf("could not find available pods: %v", err)
}
if targetRC.Spec.Replicas == readyPodCount {
// no unhealthy replicas found, no scaling required.
continue
}
scaledDownCount := int(math.Min(float64(maxCleanupCount-totalScaledDown), float64(targetRC.Spec.Replicas-readyPodCount)))
newReplicasCount := targetRC.Spec.Replicas - scaledDownCount
_, err = dc.scaleRCAndRecordEvent(targetRC, newReplicasCount, deployment)
if err != nil {
return totalScaledDown, err
}
totalScaledDown += scaledDownCount
}
return totalScaledDown, nil
}
// scaleDownOldRCsForRollingUpdate scales down old rcs when deployment strategy is "RollingUpdate".
// Need check maxUnavailable to ensure availability
func (dc *DeploymentController) scaleDownOldRCsForRollingUpdate(allRCs []*api.ReplicationController, oldRCs []*api.ReplicationController, deployment extensions.Deployment) (int, error) {
maxUnavailable, isPercent, err := util.GetIntOrPercentValue(&deployment.Spec.Strategy.RollingUpdate.MaxUnavailable)
if err != nil {
return 0, fmt.Errorf("invalid value for MaxUnavailable: %v", err)
}
if isPercent {
maxUnavailable = util.GetValueFromPercent(maxUnavailable, deployment.Spec.Replicas)
}
// Check if we can scale down.
minAvailable := deployment.Spec.Replicas - maxUnavailable
minReadySeconds := deployment.Spec.MinReadySeconds
// Find the number of ready pods.
readyPodCount, err := deploymentutil.GetAvailablePodsForRCs(dc.client, allRCs, minReadySeconds)
if err != nil {
return 0, fmt.Errorf("could not find available pods: %v", err)
}
if readyPodCount <= minAvailable {
// Cannot scale down.
return 0, nil
}
sort.Sort(controller.ControllersByCreationTimestamp(oldRCs))
totalScaledDown := 0
totalScaleDownCount := readyPodCount - minAvailable
for _, targetRC := range oldRCs {
if totalScaledDown >= totalScaleDownCount {
// No further scaling required.
break
}
@@ -857,24 +972,17 @@ func (dc *DeploymentController) reconcileOldRCs(allRCs []*api.ReplicationControl
continue
}
// Scale down.
scaleDownCount := int(math.Min(float64(targetRC.Spec.Replicas), float64(totalScaleDownCount)))
scaleDownCount := int(math.Min(float64(targetRC.Spec.Replicas), float64(totalScaleDownCount-totalScaledDown)))
newReplicasCount := targetRC.Spec.Replicas - scaleDownCount
_, err = dc.scaleRCAndRecordEvent(targetRC, newReplicasCount, deployment)
if err != nil {
return false, err
return totalScaledDown, err
}
totalScaledDown += scaleDownCount
totalScaleDownCount -= scaleDownCount
}
// Expect to see old rcs scaled down by exactly totalScaledDownCount (sum of scaleDownCount) replicas.
dKey, err = controller.KeyFunc(&deployment)
if err != nil {
return false, fmt.Errorf("Couldn't get key for deployment %#v: %v", deployment, err)
}
if expectationsCheck {
dc.podExpectations.ExpectDeletions(dKey, totalScaledDown)
}
return true, err
return totalScaledDown, nil
}
// scaleDownOldRCsForRecreate scales down old rcs when deployment strategy is "Recreate"