Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[release-v1.10] Scheduler handles overcommitted pods #820

Merged
merged 1 commit into from
Sep 8, 2023
Merged

[release-v1.10] Scheduler handles overcommitted pods #820

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
62 changes: 62 additions & 0 deletions openshift/patches/handle_overcommitted_pods.patch
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,62 @@
diff --git a/vendor/knative.dev/eventing/pkg/scheduler/state/state.go b/vendor/knative.dev/eventing/pkg/scheduler/state/state.go
index 2d5460cf8..65018e7c5 100644
--- a/vendor/knative.dev/eventing/pkg/scheduler/state/state.go
+++ b/vendor/knative.dev/eventing/pkg/scheduler/state/state.go
@@ -361,7 +361,7 @@ func (s *stateBuilder) updateFreeCapacity(free []int32, last int32, podName stri
// Assert the pod is not overcommitted
if free[ordinal] < 0 {
// This should not happen anymore. Log as an error but do not interrupt the current scheduling.
- s.logger.Errorw("pod is overcommitted", zap.String("podName", podName), zap.Int32("free", free[ordinal]))
+ s.logger.Warnw("pod is overcommitted", zap.String("podName", podName), zap.Int32("free", free[ordinal]))
}

if ordinal > last {
diff --git a/vendor/knative.dev/eventing/pkg/scheduler/statefulset/scheduler.go b/vendor/knative.dev/eventing/pkg/scheduler/statefulset/scheduler.go
index ed1defaa6..d9bcef1f8 100644
--- a/vendor/knative.dev/eventing/pkg/scheduler/statefulset/scheduler.go
+++ b/vendor/knative.dev/eventing/pkg/scheduler/statefulset/scheduler.go
@@ -272,13 +272,41 @@ func (s *StatefulSetScheduler) scheduleVPod(vpod scheduler.VPod) ([]duckv1alpha1
existingPlacements := vpod.GetPlacements()
var left int32

- // Remove unschedulable pods from placements
+ // Remove unschedulable or adjust overcommitted pods from placements
var placements []duckv1alpha1.Placement
if len(existingPlacements) > 0 {
placements = make([]duckv1alpha1.Placement, 0, len(existingPlacements))
for _, p := range existingPlacements {
- if state.IsSchedulablePod(st.OrdinalFromPodName(p.PodName)) {
- placements = append(placements, *p.DeepCopy())
+ p := p.DeepCopy()
+ ordinal := st.OrdinalFromPodName(p.PodName)
+
+ if !state.IsSchedulablePod(ordinal) {
+ continue
+ }
+
+ // Handle overcommitted pods.
+ if state.FreeCap[ordinal] < 0 {
+ // vr > free => vr: 9, overcommit 4 -> free: 0, vr: 5, pending: +4
+ // vr = free => vr: 4, overcommit 4 -> free: 0, vr: 0, pending: +4
+ // vr < free => vr: 3, overcommit 4 -> free: -1, vr: 0, pending: +3
+
+ overcommit := -state.FreeCap[ordinal]
+
+ if p.VReplicas >= overcommit {
+ state.SetFree(ordinal, 0)
+ state.Pending[vpod.GetKey()] += overcommit
+
+ p.VReplicas = p.VReplicas - overcommit
+ } else {
+ state.SetFree(ordinal, p.VReplicas-overcommit)
+ state.Pending[vpod.GetKey()] += p.VReplicas
+
+ p.VReplicas = 0
+ }
+ }
+
+ if p.VReplicas > 0 {
+ placements = append(placements, *p)
}
}
}
2 changes: 1 addition & 1 deletion vendor/knative.dev/eventing/pkg/scheduler/state/state.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -361,7 +361,7 @@ func (s *stateBuilder) updateFreeCapacity(free []int32, last int32, podName stri
// Assert the pod is not overcommitted
if free[ordinal] < 0 {
// This should not happen anymore. Log as an error but do not interrupt the current scheduling.
s.logger.Errorw("pod is overcommitted", zap.String("podName", podName), zap.Int32("free", free[ordinal]))
s.logger.Warnw("pod is overcommitted", zap.String("podName", podName), zap.Int32("free", free[ordinal]))
}

if ordinal > last {
Expand Down
34 changes: 31 additions & 3 deletions vendor/knative.dev/eventing/pkg/scheduler/statefulset/scheduler.go
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -258,13 +258,41 @@ func (s *StatefulSetScheduler) scheduleVPod(vpod scheduler.VPod) ([]duckv1alpha1
existingPlacements := vpod.GetPlacements()
var left int32

// Remove unschedulable pods from placements
// Remove unschedulable or adjust overcommitted pods from placements
var placements []duckv1alpha1.Placement
if len(existingPlacements) > 0 {
placements = make([]duckv1alpha1.Placement, 0, len(existingPlacements))
for _, p := range existingPlacements {
if state.IsSchedulablePod(st.OrdinalFromPodName(p.PodName)) {
placements = append(placements, *p.DeepCopy())
p := p.DeepCopy()
ordinal := st.OrdinalFromPodName(p.PodName)

if !state.IsSchedulablePod(ordinal) {
continue
}

// Handle overcommitted pods.
if state.FreeCap[ordinal] < 0 {
// vr > free => vr: 9, overcommit 4 -> free: 0, vr: 5, pending: +4
// vr = free => vr: 4, overcommit 4 -> free: 0, vr: 0, pending: +4
// vr < free => vr: 3, overcommit 4 -> free: -1, vr: 0, pending: +3

overcommit := -state.FreeCap[ordinal]

if p.VReplicas >= overcommit {
state.SetFree(ordinal, 0)
state.Pending[vpod.GetKey()] += overcommit

p.VReplicas = p.VReplicas - overcommit
} else {
state.SetFree(ordinal, p.VReplicas-overcommit)
state.Pending[vpod.GetKey()] += p.VReplicas

p.VReplicas = 0
}
}

if p.VReplicas > 0 {
placements = append(placements, *p)
}
}
}
Expand Down