mirror of
https://github.com/actions-runner-controller/actions-runner-controller.git
synced 2026-09-30 20:02:44 +02:00
An EphemeralRunnerSet goes Outdated when a child runner reports that the
Actions service rejected its runner spec, and the AutoscalingRunnerSet
then tears the listener down so the scale set stops taking jobs. Until
now nothing recorded which runner spec a given Outdated report was about,
so a report from a runner that predates a spec update kept the set
switched off even after the user had already fixed the spec.
Runners now carry the EphemeralRunnerSet's actionable revision as an
annotation, and Outdated runners are split into two groups:
- stale outdated: created before the currently applied revision. Their
verdict is about a spec that no longer exists, so they are deleted
and replaced by runners built from the current spec.
- outdated: created at or after the applied revision. Their verdict is
about the current spec, so they hold the set in the Outdated phase.
They are deliberately not delete-and-replaced: a fresh runner at the
same revision would report Outdated again, forever.
patchAppliedActionableRevisionStatus now recomputes the phase in both
directions against the revision being applied, because it returns from
Reconcile without reaching updateStatus. The AutoscalingRunnerSet
teardown guard additionally requires the applied revision to have caught
up with the spec revision, so it does not act on an Outdated verdict for
a spec that has already been replaced.
terminated() now allocates a fresh slice instead of appending into the
finished slice, which could alias its backing array.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
167 lines
7.7 KiB
Go
167 lines
7.7 KiB
Go
package actionsgithubcom
|
|
|
|
import (
|
|
"github.com/actions/actions-runner-controller/apis/actions.github.com/v1alpha1"
|
|
corev1 "k8s.io/api/core/v1"
|
|
apiequality "k8s.io/apimachinery/pkg/api/equality"
|
|
)
|
|
|
|
// ephemeralRunnerSetActionableSpecChanged reports whether the runner spec the
|
|
// EphemeralRunnerSet is running differs from the one derived from the
|
|
// AutoscalingRunnerSet, in a way that requires re-applying it to the runners.
|
|
//
|
|
// Semantic.DeepEqual is used rather than cmp.Equal or reflect.DeepEqual because
|
|
// it treats a nil slice/map as equal to an empty one. That matters here: most
|
|
// PodSpec collection fields carry omitempty, so a template containing an
|
|
// explicitly empty value (`env: []`) is dropped when the EphemeralRunnerSet is
|
|
// written and reads back as nil. A strict comparison would report drift on every
|
|
// single reconcile, bumping ActionableRevision each time and making the
|
|
// EphemeralRunnerSet controller delete every idle and pending runner, forever.
|
|
// Semantic also knows how to compare resource.Quantity, and unlike cmp.Equal it
|
|
// cannot panic on types with unexported fields.
|
|
func ephemeralRunnerSetActionableSpecChanged(current, desired *v1alpha1.EphemeralRunnerSet) bool {
|
|
if current == nil || desired == nil {
|
|
return current != desired
|
|
}
|
|
|
|
return !apiequality.Semantic.DeepEqual(current.Spec.EphemeralRunnerSpec, desired.Spec.EphemeralRunnerSpec)
|
|
}
|
|
|
|
func nextActionableRevision(current *v1alpha1.EphemeralRunnerSet) int64 {
|
|
if current == nil {
|
|
return 1
|
|
}
|
|
|
|
if current.Spec.ActionableRevision > current.Status.AppliedActionableRevision {
|
|
return current.Spec.ActionableRevision + 1
|
|
}
|
|
|
|
return current.Status.AppliedActionableRevision + 1
|
|
}
|
|
|
|
// ephemeralRunnerSetOutdatedForAppliedRevision reports whether the set is
|
|
// Outdated *because of the runner spec it is currently running*, which is the
|
|
// only situation in which the AutoscalingRunnerSet should tear the scale set
|
|
// down.
|
|
//
|
|
// The phase alone is not enough. Outdated is deliberately sticky: it survives in
|
|
// status while the outdated runners are collected, and it is only cleared once a
|
|
// new revision is applied. So between the moment the AutoscalingRunnerSet patches
|
|
// a new runner spec onto the set and the moment the EphemeralRunnerSet controller
|
|
// processes that patch, the set still reports Outdated for a spec that no longer
|
|
// exists. Tearing down there would discard the fix the user just applied, and the
|
|
// scale set would stay switched off until something else nudged it.
|
|
//
|
|
// Requiring the applied revision to have caught up with the spec revision closes
|
|
// that window: the verdict counts only once the set is running the current spec.
|
|
func ephemeralRunnerSetOutdatedForAppliedRevision(ephemeralRunnerSet *v1alpha1.EphemeralRunnerSet) bool {
|
|
if ephemeralRunnerSet == nil {
|
|
return false
|
|
}
|
|
|
|
return ephemeralRunnerSet.Status.Phase == v1alpha1.EphemeralRunnerSetPhaseOutdated &&
|
|
ephemeralRunnerSet.Status.AppliedActionableRevision >= ephemeralRunnerSet.Spec.ActionableRevision
|
|
}
|
|
|
|
// listenerPodSpecRequiresRecreation reports whether the live listener pod must be
|
|
// deleted and rebuilt to match the desired spec.
|
|
//
|
|
// The config secret is checked first. The pod mounts it as a volume and the
|
|
// listener parses it once at startup, so a change to the scale set URL, the TLS
|
|
// certificate, the metrics configuration or the scaler tuning only reaches the
|
|
// listener after a restart. None of that is visible in the pod spec, which
|
|
// references the secret by name and is byte-identical before and after, so the
|
|
// desired pod carries the secret's resource version as an annotation and drift
|
|
// is detected by comparing it.
|
|
//
|
|
// A pod that predates this annotation counts as drift and is recreated once, on
|
|
// the first reconcile after the controller is upgraded. Ignoring it instead is
|
|
// tempting, to avoid that rollout, but it loses updates: the reconcile that
|
|
// declines to recreate the pod goes on to patch the desired annotations onto it,
|
|
// so a config change that landed while the old controller was running is
|
|
// recorded as already applied and the listener keeps serving the configuration
|
|
// it parsed at startup, with nothing to ever correct it. The rollout is cheap by
|
|
// comparison - deleting a listener does not disturb the runners it started, and
|
|
// an upgrade normally changes the listener image and so recreates the pod
|
|
// anyway.
|
|
//
|
|
// The resource version also moves when only the secret's labels or annotations
|
|
// change, which does not affect the listener. That is accepted rather than
|
|
// worked around by hashing the secret data: those fields come from the
|
|
// AutoscalingListener spec, and a change to that spec already makes the
|
|
// AutoscalingRunnerSet controller replace the listener wholesale.
|
|
//
|
|
// DeepDerivative, not DeepEqual: the live pod carries a large number of fields
|
|
// the desired pod never sets, written by the API server and by admission
|
|
// (nodeName, dnsPolicy, schedulerName, securityContext, enableServiceLinks,
|
|
// the default tolerations, the kube-api-access-* projected volume and its mount,
|
|
// terminationMessagePath, imagePullPolicy, secret defaultMode, ...). DeepEqual
|
|
// would therefore report drift on every reconcile of every healthy listener and
|
|
// spin in a delete/create loop. It cannot be made to work by pre-populating the
|
|
// defaults either, since nodeName is scheduler-assigned and the access-token
|
|
// volume has a generated name. See TestListenerPodSpecRequiresRecreation.
|
|
//
|
|
// The cost of DeepDerivative is that it ignores empty values on the desired side,
|
|
// so a field being *removed* is invisible to it. For everything sourced from the
|
|
// user-facing template that is harmless: the AutoscalingRunnerSet controller
|
|
// compares the whole AutoscalingListener spec with cmp.Equal and deletes the
|
|
// listener outright, which takes the pod with it. Container ports are the
|
|
// exception, because they come from the --listener-metrics-addr controller flag
|
|
// rather than from any resource, so disabling metrics would otherwise leave the
|
|
// port on the pod forever. Comparing port length is enough: additions and value
|
|
// changes are already caught by DeepDerivative, only removal is blind. The
|
|
// contents are deliberately not compared, because the API server defaults
|
|
// protocol to TCP and that would reintroduce the delete/create loop.
|
|
func listenerPodSpecRequiresRecreation(current, desired *corev1.Pod) bool {
|
|
if current == nil || desired == nil {
|
|
return current != desired
|
|
}
|
|
|
|
if listenerConfigChanged(current, desired) {
|
|
return true
|
|
}
|
|
|
|
if listenerContainerPortsRemoved(current, desired) {
|
|
return true
|
|
}
|
|
|
|
return !apiequality.Semantic.DeepDerivative(desired.Spec, current.Spec)
|
|
}
|
|
|
|
func listenerConfigChanged(current, desired *corev1.Pod) bool {
|
|
desiredVersion := desired.Annotations[AnnotationKeyListenerConfigResourceVersion]
|
|
if desiredVersion == "" {
|
|
// Nothing to compare against, so there is nothing this check can say.
|
|
// The desired pod is always built from a secret read back from the API
|
|
// server, so this only happens in tests.
|
|
return false
|
|
}
|
|
|
|
return current.Annotations[AnnotationKeyListenerConfigResourceVersion] != desiredVersion
|
|
}
|
|
|
|
func listenerContainerPortsRemoved(current, desired *corev1.Pod) bool {
|
|
for i := range desired.Spec.Containers {
|
|
desiredContainer := &desired.Spec.Containers[i]
|
|
currentContainer := findContainerByName(current.Spec.Containers, desiredContainer.Name)
|
|
if currentContainer == nil {
|
|
// A container the live pod does not have at all is drift that
|
|
// DeepDerivative already reports; nothing to decide here.
|
|
continue
|
|
}
|
|
if len(desiredContainer.Ports) < len(currentContainer.Ports) {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
func findContainerByName(containers []corev1.Container, name string) *corev1.Container {
|
|
for i := range containers {
|
|
if containers[i].Name == name {
|
|
return &containers[i]
|
|
}
|
|
}
|
|
return nil
|
|
}
|