Fix overscaling when the controller is much faster then the listener (#3371)

Co-authored-by: Francesco Renzi <rentziass@gmail.com>
This commit is contained in:
Nikola Jokic
2024-03-20 15:36:12 +01:00
committed by GitHub
parent 46cfbb6ec7
commit 7a643a5107
16 changed files with 262 additions and 158 deletions

View File

@@ -22,6 +22,7 @@ import (
"fmt"
"net/http"
"sort"
"strconv"
"strings"
"github.com/actions/actions-runner-controller/apis/actions.github.com/v1alpha1"
@@ -156,14 +157,14 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
return ctrl.Result{}, err
}
pendingEphemeralRunners, runningEphemeralRunners, finishedEphemeralRunners, failedEphemeralRunners, deletingEphemeralRunners := categorizeEphemeralRunners(ephemeralRunnerList)
ephemeralRunnerState := newEphemeralRunnerState(ephemeralRunnerList)
log.Info("Ephemeral runner counts",
"pending", len(pendingEphemeralRunners),
"running", len(runningEphemeralRunners),
"finished", len(finishedEphemeralRunners),
"failed", len(failedEphemeralRunners),
"deleting", len(deletingEphemeralRunners),
"pending", len(ephemeralRunnerState.pending),
"running", len(ephemeralRunnerState.running),
"finished", len(ephemeralRunnerState.finished),
"failed", len(ephemeralRunnerState.failed),
"deleting", len(ephemeralRunnerState.deleting),
)
if r.PublishMetrics {
@@ -183,54 +184,52 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
Organization: parsedURL.Organization,
Enterprise: parsedURL.Enterprise,
},
len(pendingEphemeralRunners),
len(runningEphemeralRunners),
len(failedEphemeralRunners),
len(ephemeralRunnerState.pending),
len(ephemeralRunnerState.running),
len(ephemeralRunnerState.failed),
)
}
// cleanup finished runners and proceed
var errs []error
for i := range finishedEphemeralRunners {
log.Info("Deleting finished ephemeral runner", "name", finishedEphemeralRunners[i].Name)
if err := r.Delete(ctx, finishedEphemeralRunners[i]); err != nil {
if !kerrors.IsNotFound(err) {
errs = append(errs, err)
total := ephemeralRunnerState.scaleTotal()
if ephemeralRunnerSet.Spec.PatchID == 0 || ephemeralRunnerSet.Spec.PatchID != ephemeralRunnerState.latestPatchID {
defer func() {
if err := r.cleanupFinishedEphemeralRunners(ctx, ephemeralRunnerState.finished, log); err != nil {
log.Error(err, "failed to cleanup finished ephemeral runners")
}
}
}
}()
if len(errs) > 0 {
mergedErrs := multierr.Combine(errs...)
log.Error(mergedErrs, "Failed to delete finished ephemeral runners")
return ctrl.Result{}, mergedErrs
}
log.Info("Scaling comparison", "current", total, "desired", ephemeralRunnerSet.Spec.Replicas)
switch {
case total < ephemeralRunnerSet.Spec.Replicas: // Handle scale up
count := ephemeralRunnerSet.Spec.Replicas - total
log.Info("Creating new ephemeral runners (scale up)", "count", count)
if err := r.createEphemeralRunners(ctx, ephemeralRunnerSet, count, log); err != nil {
log.Error(err, "failed to make ephemeral runner")
return ctrl.Result{}, err
}
total := len(pendingEphemeralRunners) + len(runningEphemeralRunners) + len(failedEphemeralRunners)
log.Info("Scaling comparison", "current", total, "desired", ephemeralRunnerSet.Spec.Replicas)
switch {
case total < ephemeralRunnerSet.Spec.Replicas: // Handle scale up
count := ephemeralRunnerSet.Spec.Replicas - total
log.Info("Creating new ephemeral runners (scale up)", "count", count)
if err := r.createEphemeralRunners(ctx, ephemeralRunnerSet, count, log); err != nil {
log.Error(err, "failed to make ephemeral runner")
return ctrl.Result{}, err
}
case total > ephemeralRunnerSet.Spec.Replicas: // Handle scale down scenario.
count := total - ephemeralRunnerSet.Spec.Replicas
log.Info("Deleting ephemeral runners (scale down)", "count", count)
if err := r.deleteIdleEphemeralRunners(ctx, ephemeralRunnerSet, pendingEphemeralRunners, runningEphemeralRunners, count, log); err != nil {
log.Error(err, "failed to delete idle runners")
return ctrl.Result{}, err
case total > ephemeralRunnerSet.Spec.Replicas: // Handle scale down scenario.
count := total - ephemeralRunnerSet.Spec.Replicas
log.Info("Deleting ephemeral runners (scale down)", "count", count)
if err := r.deleteIdleEphemeralRunners(
ctx,
ephemeralRunnerSet,
ephemeralRunnerState.pending,
ephemeralRunnerState.running,
count,
log,
); err != nil {
log.Error(err, "failed to delete idle runners")
return ctrl.Result{}, err
}
}
}
desiredStatus := v1alpha1.EphemeralRunnerSetStatus{
CurrentReplicas: total,
PendingEphemeralRunners: len(pendingEphemeralRunners),
RunningEphemeralRunners: len(runningEphemeralRunners),
FailedEphemeralRunners: len(failedEphemeralRunners),
PendingEphemeralRunners: len(ephemeralRunnerState.pending),
RunningEphemeralRunners: len(ephemeralRunnerState.running),
FailedEphemeralRunners: len(ephemeralRunnerState.failed),
}
// Update the status if needed.
@@ -247,6 +246,21 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
return ctrl.Result{}, nil
}
func (r *EphemeralRunnerSetReconciler) cleanupFinishedEphemeralRunners(ctx context.Context, finishedEphemeralRunners []*v1alpha1.EphemeralRunner, log logr.Logger) error {
// cleanup finished runners and proceed
var errs []error
for i := range finishedEphemeralRunners {
log.Info("Deleting finished ephemeral runner", "name", finishedEphemeralRunners[i].Name)
if err := r.Delete(ctx, finishedEphemeralRunners[i]); err != nil {
if !kerrors.IsNotFound(err) {
errs = append(errs, err)
}
}
}
return multierr.Combine(errs...)
}
func (r *EphemeralRunnerSetReconciler) cleanUpProxySecret(ctx context.Context, ephemeralRunnerSet *v1alpha1.EphemeralRunnerSet, log logr.Logger) error {
if ephemeralRunnerSet.Spec.EphemeralRunnerSpec.Proxy == nil {
return nil
@@ -284,19 +298,19 @@ func (r *EphemeralRunnerSetReconciler) cleanUpEphemeralRunners(ctx context.Conte
return true, nil
}
pendingEphemeralRunners, runningEphemeralRunners, finishedEphemeralRunners, failedEphemeralRunners, deletingEphemeralRunners := categorizeEphemeralRunners(ephemeralRunnerList)
ephemeralRunnerState := newEphemeralRunnerState(ephemeralRunnerList)
log.Info("Clean up runner counts",
"pending", len(pendingEphemeralRunners),
"running", len(runningEphemeralRunners),
"finished", len(finishedEphemeralRunners),
"failed", len(failedEphemeralRunners),
"deleting", len(deletingEphemeralRunners),
"pending", len(ephemeralRunnerState.pending),
"running", len(ephemeralRunnerState.running),
"finished", len(ephemeralRunnerState.finished),
"failed", len(ephemeralRunnerState.failed),
"deleting", len(ephemeralRunnerState.deleting),
)
log.Info("Cleanup finished or failed ephemeral runners")
var errs []error
for _, ephemeralRunner := range append(finishedEphemeralRunners, failedEphemeralRunners...) {
for _, ephemeralRunner := range append(ephemeralRunnerState.finished, ephemeralRunnerState.failed...) {
log.Info("Deleting ephemeral runner", "name", ephemeralRunner.Name)
if err := r.Delete(ctx, ephemeralRunner); err != nil && !kerrors.IsNotFound(err) {
errs = append(errs, err)
@@ -310,7 +324,7 @@ func (r *EphemeralRunnerSetReconciler) cleanUpEphemeralRunners(ctx context.Conte
}
// avoid fetching the client if we have nothing left to do
if len(runningEphemeralRunners) == 0 && len(pendingEphemeralRunners) == 0 {
if len(ephemeralRunnerState.running) == 0 && len(ephemeralRunnerState.pending) == 0 {
return false, nil
}
@@ -321,7 +335,7 @@ func (r *EphemeralRunnerSetReconciler) cleanUpEphemeralRunners(ctx context.Conte
log.Info("Cleanup pending or running ephemeral runners")
errs = errs[0:0]
for _, ephemeralRunner := range append(pendingEphemeralRunners, runningEphemeralRunners...) {
for _, ephemeralRunner := range append(ephemeralRunnerState.pending, ephemeralRunnerState.running...) {
log.Info("Removing the ephemeral runner from the service", "name", ephemeralRunner.Name)
_, err := r.deleteEphemeralRunnerWithActionsClient(ctx, ephemeralRunner, actionsClient, log)
if err != nil {
@@ -427,12 +441,13 @@ func (r *EphemeralRunnerSetReconciler) deleteIdleEphemeralRunners(ctx context.Co
deletedCount := 0
for runners.next() {
ephemeralRunner := runners.object()
if ephemeralRunner.Status.RunnerId == 0 {
isDone := ephemeralRunner.IsDone()
if !isDone && ephemeralRunner.Status.RunnerId == 0 {
log.Info("Skipping ephemeral runner since it is not registered yet", "name", ephemeralRunner.Name)
continue
}
if ephemeralRunner.Status.JobRequestId > 0 {
if !isDone && ephemeralRunner.Status.JobRequestId > 0 {
log.Info("Skipping ephemeral runner since it is running a job", "name", ephemeralRunner.Name, "jobRequestId", ephemeralRunner.Status.JobRequestId)
continue
}
@@ -580,16 +595,22 @@ type ephemeralRunnerStepper struct {
index int
}
func newEphemeralRunnerStepper(pending, running []*v1alpha1.EphemeralRunner) *ephemeralRunnerStepper {
sort.Slice(pending, func(i, j int) bool {
return pending[i].GetCreationTimestamp().Time.Before(pending[j].GetCreationTimestamp().Time)
})
sort.Slice(running, func(i, j int) bool {
return running[i].GetCreationTimestamp().Time.Before(running[j].GetCreationTimestamp().Time)
func newEphemeralRunnerStepper(primary []*v1alpha1.EphemeralRunner, othersOrdered ...[]*v1alpha1.EphemeralRunner) *ephemeralRunnerStepper {
sort.Slice(primary, func(i, j int) bool {
return primary[i].GetCreationTimestamp().Time.Before(primary[j].GetCreationTimestamp().Time)
})
for _, bucket := range othersOrdered {
sort.Slice(bucket, func(i, j int) bool {
return bucket[i].GetCreationTimestamp().Time.Before(bucket[j].GetCreationTimestamp().Time)
})
}
for _, bucket := range othersOrdered {
primary = append(primary, bucket...)
}
return &ephemeralRunnerStepper{
items: append(pending, running...),
items: primary,
index: -1,
}
}
@@ -613,28 +634,48 @@ func (s *ephemeralRunnerStepper) len() int {
return len(s.items)
}
func categorizeEphemeralRunners(ephemeralRunnerList *v1alpha1.EphemeralRunnerList) (pendingEphemeralRunners, runningEphemeralRunners, finishedEphemeralRunners, failedEphemeralRunners, deletingEphemeralRunners []*v1alpha1.EphemeralRunner) {
type ephemeralRunnerState struct {
pending []*v1alpha1.EphemeralRunner
running []*v1alpha1.EphemeralRunner
finished []*v1alpha1.EphemeralRunner
failed []*v1alpha1.EphemeralRunner
deleting []*v1alpha1.EphemeralRunner
latestPatchID int
}
func newEphemeralRunnerState(ephemeralRunnerList *v1alpha1.EphemeralRunnerList) *ephemeralRunnerState {
var ephemeralRunnerState ephemeralRunnerState
for i := range ephemeralRunnerList.Items {
r := &ephemeralRunnerList.Items[i]
patchID, err := strconv.Atoi(r.Annotations[AnnotationKeyPatchID])
if err == nil && patchID > ephemeralRunnerState.latestPatchID {
ephemeralRunnerState.latestPatchID = patchID
}
if !r.ObjectMeta.DeletionTimestamp.IsZero() {
deletingEphemeralRunners = append(deletingEphemeralRunners, r)
ephemeralRunnerState.deleting = append(ephemeralRunnerState.deleting, r)
continue
}
switch r.Status.Phase {
case corev1.PodRunning:
runningEphemeralRunners = append(runningEphemeralRunners, r)
ephemeralRunnerState.running = append(ephemeralRunnerState.running, r)
case corev1.PodSucceeded:
finishedEphemeralRunners = append(finishedEphemeralRunners, r)
ephemeralRunnerState.finished = append(ephemeralRunnerState.finished, r)
case corev1.PodFailed:
failedEphemeralRunners = append(failedEphemeralRunners, r)
ephemeralRunnerState.failed = append(ephemeralRunnerState.failed, r)
default:
// Pending or no phase should be considered as pending.
//
// If field is not set, that means that the EphemeralRunner
// did not yet have chance to update the Status.Phase field.
pendingEphemeralRunners = append(pendingEphemeralRunners, r)
ephemeralRunnerState.pending = append(ephemeralRunnerState.pending, r)
}
}
return
return &ephemeralRunnerState
}
func (s *ephemeralRunnerState) scaleTotal() int {
return len(s.pending) + len(s.running) + len(s.failed)
}