Take the reconcile snapshot lazily, right before the first mutation

Reconcilers deep copied the object they had just fetched on every single
reconcile, purely so a merge patch could be computed on the rare pass
that actually changes something. The copy is a full recursive walk and
allocation of the object, and the overwhelming majority of reconciles
throw it away untouched.

Introduce lazyCopy, which takes the snapshot on the first call to
Mutate and hands back the live object. Because Mutate is the only way
to reach the object, the snapshot cannot be taken after the mutation it
is supposed to be diffed against, which is the way this optimization is
usually gotten wrong.

Apply it to the four Reconcile entry points, and move the two
EphemeralRunnerSet status copies inside the branch that patches, so they
are only paid for when the status really changed.

While here, drop the short circuit in the EphemeralRunner finalizer
block: `addedFinalizers || AddFinalizer(...)` skipped adding the actions
finalizer whenever the first finalizer was added.

Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
This commit is contained in:
Nikola Jokic
2026-09-11 11:32:03 +02:00
co-authored by Copilot App
parent f7dd0d2d01
commit a7ce96b76c
6 changed files with 179 additions and 37 deletions
@@ -87,7 +87,7 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
if err := r.Get(ctx, req.NamespacedName, &ephemeralRunnerSet); err != nil {
return ctrl.Result{}, client.IgnoreNotFound(err)
}
original := ephemeralRunnerSet.DeepCopy()
runnerSet := newLazyCopy(&ephemeralRunnerSet)
// Requested deletion does not need reconciled.
if !ephemeralRunnerSet.DeletionTimestamp.IsZero() {
@@ -117,8 +117,8 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
}
log.Info("Removing finalizer")
if controllerutil.RemoveFinalizer(&ephemeralRunnerSet, EphemeralRunnerSetFinalizerName) {
if err := r.Patch(ctx, &ephemeralRunnerSet, client.MergeFrom(original)); err != nil {
if controllerutil.RemoveFinalizer(runnerSet.Mutate(), EphemeralRunnerSetFinalizerName) {
if err := r.Patch(ctx, &ephemeralRunnerSet, runnerSet.MergeFrom()); err != nil {
log.Error(err, "Failed to update ephemeral runner set with removed finalizer")
return ctrl.Result{}, err
}
@@ -130,9 +130,10 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
}
// Add finalizer if not present
if controllerutil.AddFinalizer(&ephemeralRunnerSet, EphemeralRunnerSetFinalizerName) {
if !controllerutil.ContainsFinalizer(&ephemeralRunnerSet, EphemeralRunnerSetFinalizerName) {
controllerutil.AddFinalizer(runnerSet.Mutate(), EphemeralRunnerSetFinalizerName)
log.Info("Adding finalizer")
if err := r.Patch(ctx, &ephemeralRunnerSet, client.MergeFrom(original)); err != nil {
if err := r.Patch(ctx, &ephemeralRunnerSet, runnerSet.MergeFrom()); err != nil {
log.Error(err, "Failed to update ephemeral runner set with new finalizer")
return ctrl.Result{}, err
}
@@ -353,14 +354,16 @@ func (r *EphemeralRunnerSetReconciler) patchAppliedActionableRevisionStatus(ctx
return err
}
original := latest.DeepCopy()
// Build the desired status as a value so no copy of the object is taken
// on the common path where the status already matches.
desiredStatus := latest.Status
// Only an advance means the idle and pending runners were just deleted and
// the listener restarted. Guarding both writes on it keeps this callable
// as a plain "make sure status reflects revision N" without disturbing a
// marker that still describes the live patch sequence.
if latest.Status.AppliedActionableRevision < targetAppliedRevision {
latest.Status.AppliedActionableRevision = targetAppliedRevision
desiredStatus.AppliedActionableRevision = targetAppliedRevision
// The marker records a patch ID from the sequence that was current
// before this spec change. Applying a new revision deletes the idle and
@@ -370,7 +373,7 @@ func (r *EphemeralRunnerSetReconciler) patchAppliedActionableRevisionStatus(ctx
// through every integer. It therefore passes through a leftover marker
// value with near-certainty, and would suppress the very scale up that
// rebuilds the pool.
latest.Status.FinishedRunnerCleanupPatchID = 0
desiredStatus.FinishedRunnerCleanupPatchID = 0
}
ephemeralRunnerList := new(v1alpha1.EphemeralRunnerList)
@@ -408,17 +411,20 @@ func (r *EphemeralRunnerSetReconciler) patchAppliedActionableRevisionStatus(ctx
// outdated runners from the cleanup path, and a stale Outdated would keep
// the set switched off after the spec that caused it was replaced.
if len(state.outdated) > 0 {
latest.Status.Phase = v1alpha1.EphemeralRunnerSetPhaseOutdated
desiredStatus.Phase = v1alpha1.EphemeralRunnerSetPhaseOutdated
} else {
latest.Status.Phase = v1alpha1.EphemeralRunnerSetPhaseRunning
desiredStatus.Phase = v1alpha1.EphemeralRunnerSetPhaseRunning
}
// Checked after every field above has been set, so that clearing the
// marker alone is still enough to issue the patch.
if original.Status == latest.Status {
if latest.Status == desiredStatus {
return nil
}
original := latest.DeepCopy()
latest.Status = desiredStatus
return r.Status().Patch(ctx, &latest, client.MergeFromWithOptions(original, client.MergeFromWithOptimisticLock{}))
})
}
@@ -534,7 +540,6 @@ func (r *EphemeralRunnerSetReconciler) patchFinishedRunnerCleanupPatchIDStatus(c
}
func (r *EphemeralRunnerSetReconciler) updateStatus(ctx context.Context, ephemeralRunnerSet *v1alpha1.EphemeralRunnerSet, state *ephemeralRunnersByState, log logr.Logger) error {
original := ephemeralRunnerSet.DeepCopy()
var phase v1alpha1.EphemeralRunnerSetPhase
switch {
case len(state.outdated) > 0:
@@ -552,6 +557,7 @@ func (r *EphemeralRunnerSetReconciler) updateStatus(ctx context.Context, ephemer
// Update the status if needed.
if ephemeralRunnerSet.Status != desiredStatus {
original := ephemeralRunnerSet.DeepCopy()
ephemeralRunnerSet.Status = desiredStatus
if err := r.Status().Patch(ctx, ephemeralRunnerSet, client.MergeFrom(original)); err != nil {
log.Error(err, "Failed to update EphemeralRunnerSet status")