extending ephemeral runner statuses

.github/workflows/arc-publish-chart.yaml

@@ -63,7 +63,7 @@ jobs:
           python-version: "3.11"
 
       - name: Set up chart-testing
-        uses: helm/chart-testing-action@6ec842c01de15ebb84c8627d2744a0c2f2755c9f
+        uses: helm/chart-testing-action@0d28d3144d3a25ea2cc349d6e59901c4ff469b3b
 
       - name: Run chart-testing (list-changed)
         id: list-changed
@@ -79,7 +79,7 @@ jobs:
 
       - name: Create kind cluster
         if: steps.list-changed.outputs.changed == 'true'
-        uses: helm/kind-action@92086f6be054225fa813e0a4b13787fc9088faab
+        uses: helm/kind-action@a1b0e391336a6ee6713a0583f8c6240d70863de3
 
       # We need cert-manager already installed in the cluster because we assume the CRDs exist
       - name: Install cert-manager

.github/workflows/arc-release-runners.yaml

@@ -19,7 +19,7 @@ env:
   PUSH_TO_REGISTRIES: true
   TARGET_ORG: actions-runner-controller
   TARGET_WORKFLOW: release-runners.yaml
-  DOCKER_VERSION: 28.0.4
+  DOCKER_VERSION: 24.0.7
 
 concurrency:
   group: ${{ github.workflow }}

.github/workflows/arc-validate-chart.yaml

@@ -55,7 +55,7 @@ jobs:
           python-version: "3.11"
 
       - name: Set up chart-testing
-        uses: helm/chart-testing-action@6ec842c01de15ebb84c8627d2744a0c2f2755c9f
+        uses: helm/chart-testing-action@0d28d3144d3a25ea2cc349d6e59901c4ff469b3b
 
       - name: Run chart-testing (list-changed)
         id: list-changed
@@ -70,7 +70,7 @@ jobs:
           ct lint --config charts/.ci/ct-config.yaml
 
       - name: Create kind cluster
-        uses: helm/kind-action@92086f6be054225fa813e0a4b13787fc9088faab
+        uses: helm/kind-action@a1b0e391336a6ee6713a0583f8c6240d70863de3
         if: steps.list-changed.outputs.changed == 'true'
 
       # We need cert-manager already installed in the cluster because we assume the CRDs exist

.github/workflows/gha-e2e-tests.yaml

@@ -984,7 +984,7 @@ jobs:
               echo "5 pods are up!"
               break
             fi
-            if [[ "$count" -ge 30 ]]; then
+            if [[ "$count" -ge 12 ]]; then
               echo "Timeout waiting for 5 pods to be created"
               exit 1
             fi

.github/workflows/gha-publish-chart.yaml

@@ -72,7 +72,7 @@ jobs:
           echo "repository_owner=$(echo ${{ github.repository_owner }} | tr '[:upper:]' '[:lower:]')" >> $GITHUB_OUTPUT
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130
+        uses: docker/setup-qemu-action@29109295f81e9208d7d86ff1c6c12d2833863392
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@e468171a9de216ec08956ac3ada2f0791b6bd435

.github/workflows/gha-validate-chart.yaml

@@ -51,7 +51,7 @@ jobs:
           python-version: "3.11"
 
       - name: Set up chart-testing
-        uses: helm/chart-testing-action@6ec842c01de15ebb84c8627d2744a0c2f2755c9f
+        uses: helm/chart-testing-action@0d28d3144d3a25ea2cc349d6e59901c4ff469b3b
 
       - name: Run chart-testing (list-changed)
         id: list-changed
@@ -88,7 +88,7 @@ jobs:
           cache-to: type=gha,mode=max
 
       - name: Create kind cluster
-        uses: helm/kind-action@92086f6be054225fa813e0a4b13787fc9088faab
+        uses: helm/kind-action@a1b0e391336a6ee6713a0583f8c6240d70863de3
         if: steps.list-changed.outputs.changed == 'true'
         with:
           cluster_name: chart-testing

.github/workflows/global-publish-canary.yaml

@@ -110,7 +110,7 @@ jobs:
           echo "repository_owner=$(echo ${{ github.repository_owner }} | tr '[:upper:]' '[:lower:]')" >> $GITHUB_OUTPUT
 
       - name: Set up QEMU
-        uses: docker/setup-qemu-action@c7c53464625b32c7a7e944ae62b3e17d2b600130
+        uses: docker/setup-qemu-action@29109295f81e9208d7d86ff1c6c12d2833863392
 
       - name: Set up Docker Buildx
         uses: docker/setup-buildx-action@e468171a9de216ec08956ac3ada2f0791b6bd435

.github/workflows/go.yaml

@@ -48,7 +48,7 @@ jobs:
           go-version-file: "go.mod"
           cache: false
       - name: golangci-lint
-        uses: golangci/golangci-lint-action@0a35821d5c230e903fcfe077583637dea1b27b47
+        uses: golangci/golangci-lint-action@4afd733a84b1f43292c63897423277bb7f4313a9
         with:
           only-new-issues: true
           version: v2.5.0
@@ -78,7 +78,7 @@ jobs:
         run: git diff --exit-code
       - name: Install kubebuilder
         run: |
-          curl -D headers.txt -fsL "https://storage.googleapis.com/kubebuilder-tools/kubebuilder-tools-1.30.0-linux-amd64.tar.gz" -o kubebuilder-tools
+          curl -D headers.txt -fsL "https://storage.googleapis.com/kubebuilder-tools/kubebuilder-tools-1.26.1-linux-amd64.tar.gz" -o kubebuilder-tools
           echo "$(grep -i etag headers.txt -m 1 | cut -d'"' -f2) kubebuilder-tools" > sum
           md5sum -c sum
           tar -zvxf kubebuilder-tools

apis/actions.github.com/v1alpha1/ephemeralrunner_types.go

@@ -48,7 +48,7 @@ type EphemeralRunner struct {
 }
 
 func (er *EphemeralRunner) IsDone() bool {
-	return er.Status.Phase == corev1.PodSucceeded || er.Status.Phase == corev1.PodFailed
+	return er.Status.Phase == EphemeralRunnerSucceeded || er.Status.Phase == EphemeralRunnerFailed
 }
 
 func (er *EphemeralRunner) HasJob() bool {
@@ -125,6 +125,40 @@ type EphemeralRunnerSpec struct {
 	corev1.PodTemplateSpec `json:",inline"`
 }
 
+// EphemeralRunnerPhase is a label for the condition of an EphemeralRunner at the current time.
+// +kubebuilder:validation:Enum=Pending;Running;Restarting;Succeeded;Failed;Aborted
+type EphemeralRunnerPhase string
+
+const (
+	// EphemeralRunnerPending is the stage where the ephemeral runner is about to start.
+	EphemeralRunnerPending EphemeralRunnerPhase = "Pending"
+	// EphemeralRunnerRunning is the stage where the ephemeral runner is running and ready to accept the job.
+	EphemeralRunnerRunning EphemeralRunnerPhase = "Running"
+	// EphemeralRunnerRestarting is the stage where the ephemeral runner pod stopped, so the ephemeral runner should restart it
+	EphemeralRunnerRestarting EphemeralRunnerPhase = "Restarting"
+	// EphemeralRunnerSucceeded is the stage where the ephemeral runner finished running and exited with exit code 0.
+	EphemeralRunnerSucceeded EphemeralRunnerPhase = "Succeeded"
+	// EphemeralRunnerFailed means that the ephemeral runner finished running and exited with a non-zero exit code.
+	EphemeralRunnerFailed EphemeralRunnerPhase = "Failed"
+	// EphemeralRunnerAborted means that the ephemeral runner failed to start due to unrecoverable failure, and will be left as is for manual inspection.
+	EphemeralRunnerAborted EphemeralRunnerPhase = "Aborted"
+)
+
+func EphemeralRunnerPhaseFromPodPhase(podPhase corev1.PodPhase) EphemeralRunnerPhase {
+	switch podPhase {
+	case corev1.PodPending:
+		return EphemeralRunnerPending
+	case corev1.PodRunning:
+		return EphemeralRunnerRunning
+	case corev1.PodSucceeded:
+		return EphemeralRunnerSucceeded
+	case corev1.PodFailed:
+		return EphemeralRunnerFailed
+	default:
+		return EphemeralRunnerPending
+	}
+}
+
 // EphemeralRunnerStatus defines the observed state of EphemeralRunner
 type EphemeralRunnerStatus struct {
 	// Turns true only if the runner is online.
@@ -140,7 +174,7 @@ type EphemeralRunnerStatus struct {
 	// The PodSucceded phase should be set only when confirmed that EphemeralRunner
 	// actually executed the job and has been removed from the service.
 	// +optional
-	Phase corev1.PodPhase `json:"phase,omitempty"`
+	Phase EphemeralRunnerPhase `json:"phase,omitempty"`
 	// +optional
 	Reason string `json:"reason,omitempty"`
 	// +optional

charts/gha-runner-scale-set-controller/crds/actions.github.com_ephemeralrunners.yaml

@@ -8257,6 +8257,13 @@ spec:
 
                     The PodSucceded phase should be set only when confirmed that EphemeralRunner
                     actually executed the job and has been removed from the service.
+                  enum:
+                    - Pending
+                    - Running
+                    - Restarting
+                    - Succeeded
+                    - Failed
+                    - Aborted
                   type: string
                 ready:
                   description: Turns true only if the runner is online.

config/crd/bases/actions.github.com_ephemeralrunners.yaml

@@ -8257,6 +8257,13 @@ spec:
 
                     The PodSucceded phase should be set only when confirmed that EphemeralRunner
                     actually executed the job and has been removed from the service.
+                  enum:
+                    - Pending
+                    - Running
+                    - Restarting
+                    - Succeeded
+                    - Failed
+                    - Aborted
                   type: string
                 ready:
                   description: Turns true only if the runner is online.

controllers/actions.github.com/ephemeralrunner_controller.go

@@ -22,7 +22,6 @@ import (
 	"fmt"
 	"net/http"
 	"strconv"
-	"strings"
 	"time"
 
 	"github.com/actions/actions-runner-controller/apis/actions.github.com/v1alpha1"
@@ -283,37 +282,10 @@ func (r *EphemeralRunnerReconciler) Reconcile(ctx context.Context, req ctrl.Requ
 		case kerrors.IsAlreadyExists(err):
 			log.Info("Runner pod already exists. Waiting for the pod event to be received")
 			return ctrl.Result{Requeue: true, RequeueAfter: 5 * time.Second}, nil
-		case kerrors.IsInvalid(err):
+		case kerrors.IsInvalid(err) || kerrors.IsForbidden(err):
 			log.Error(err, "Failed to create a pod due to unrecoverable failure")
 			errMessage := fmt.Sprintf("Failed to create the pod: %v", err)
-			if err := r.markAsFailed(ctx, ephemeralRunner, errMessage, ReasonInvalidPodFailure, log); err != nil {
+			if err := r.markAsAborted(ctx, ephemeralRunner, errMessage, ReasonInvalidPodFailure, log); err != nil {
-				log.Error(err, "Failed to set ephemeral runner to phase Failed")
-				return ctrl.Result{}, err
-			}
-			return ctrl.Result{}, nil
-		case kerrors.IsForbidden(err):
-			if status, ok := err.(kerrors.APIStatus); ok || errors.As(err, &status) {
-				isResourceQuotaExceeded := strings.Contains(status.Status().Message, "exceeded quota:")
-				isAboutToExpire := ephemeralRunner.CreationTimestamp.Time.Add(10 * time.Minute).Before(time.Now())
-				switch {
-				case isResourceQuotaExceeded && isAboutToExpire:
-					log.Error(err, "Failed to create a pod due to resource quota exceeded and the ephemeral runner is about to expire; re-creating the ephemeral runner")
-					if err := r.Delete(ctx, ephemeralRunner); err != nil {
-						log.Error(err, "Failed to delete the ephemeral runner")
-						return ctrl.Result{}, err
-					}
-					return ctrl.Result{}, nil
-				case isResourceQuotaExceeded:
-					log.Error(err, "Resource quota is exceeded; requeue in 30s to retry pod creation")
-					return ctrl.Result{RequeueAfter: 30 * time.Second}, nil
-				default:
-					// other forbidden errors
-					// fallthrough to the default handling below
-				}
-			}
-			log.Error(err, "Failed to create a pod due to unrecoverable failure")
-			errMessage := fmt.Sprintf("Failed to create the pod: %v", err)
-			if err := r.markAsFailed(ctx, ephemeralRunner, errMessage, ReasonInvalidPodFailure, log); err != nil {
 				log.Error(err, "Failed to set ephemeral runner to phase Failed")
 				return ctrl.Result{}, err
 			}
@@ -539,10 +511,12 @@ func (r *EphemeralRunnerReconciler) cleanupRunnerLinkedSecrets(ctx context.Conte
 	return errors.Join(errs...)
 }
 
-func (r *EphemeralRunnerReconciler) markAsFailed(ctx context.Context, ephemeralRunner *v1alpha1.EphemeralRunner, errMessage string, reason string, log logr.Logger) error {
+// markAsAborted updates the ephemeral runner status to aborted and stops the reconciliation.
+// This runner is left in aborted state and won't be deleted until someone manually does that.
+func (r *EphemeralRunnerReconciler) markAsAborted(ctx context.Context, ephemeralRunner *v1alpha1.EphemeralRunner, errMessage string, reason string, log logr.Logger) error {
 	log.Info("Updating ephemeral runner status to Failed")
 	if err := patchSubResource(ctx, r.Status(), ephemeralRunner, func(obj *v1alpha1.EphemeralRunner) {
-		obj.Status.Phase = corev1.PodFailed
+		obj.Status.Phase = v1alpha1.EphemeralRunnerAborted
 		obj.Status.Reason = reason
 		obj.Status.Message = errMessage
 	}); err != nil {
@@ -573,6 +547,7 @@ func (r *EphemeralRunnerReconciler) deletePodAsFailed(ctx context.Context, ephem
 		if obj.Status.Failures == nil {
 			obj.Status.Failures = make(map[string]metav1.Time)
 		}
+		obj.Status.Phase = v1alpha1.EphemeralRunnerRestarting
 		obj.Status.Failures[string(pod.UID)] = metav1.Now()
 		obj.Status.Ready = false
 		obj.Status.Reason = pod.Status.Reason
@@ -762,7 +737,8 @@ func (r *EphemeralRunnerReconciler) updateRunStatusFromPod(ctx context.Context,
 		}
 	}
 
-	phaseChanged := ephemeralRunner.Status.Phase != pod.Status.Phase
+	newPhase := v1alpha1.EphemeralRunnerPhaseFromPodPhase(pod.Status.Phase)
+	phaseChanged := ephemeralRunner.Status.Phase != newPhase
 	readyChanged := ready != ephemeralRunner.Status.Ready
 
 	if !phaseChanged && !readyChanged {
@@ -777,7 +753,7 @@ func (r *EphemeralRunnerReconciler) updateRunStatusFromPod(ctx context.Context,
 		"ready", ready,
 	)
 	err := patchSubResource(ctx, r.Status(), ephemeralRunner, func(obj *v1alpha1.EphemeralRunner) {
-		obj.Status.Phase = pod.Status.Phase
+		obj.Status.Phase = newPhase
 		obj.Status.Ready = ready
 		obj.Status.Reason = pod.Status.Reason
 		obj.Status.Message = pod.Status.Message

controllers/actions.github.com/ephemeralrunner_controller_test.go

@@ -270,7 +270,7 @@ var _ = Describe("EphemeralRunner", func() {
 
 			updated := new(v1alpha1.EphemeralRunner)
 			Eventually(
-				func() (corev1.PodPhase, error) {
+				func() (v1alpha1.EphemeralRunnerPhase, error) {
 					err := k8sClient.Get(
 						ctx,
 						client.ObjectKey{Name: invalideEphemeralRunner.Name, Namespace: invalideEphemeralRunner.Namespace},
@@ -283,7 +283,7 @@ var _ = Describe("EphemeralRunner", func() {
 				},
 				ephemeralRunnerTimeout,
 				ephemeralRunnerInterval,
-			).Should(BeEquivalentTo(corev1.PodFailed))
+			).Should(BeEquivalentTo(v1alpha1.EphemeralRunnerAborted))
 
 			Expect(updated.Status.Reason).Should(Equal("InvalidPod"))
 			Expect(updated.Status.Message).Should(Equal("Failed to create the pod: pods \"invalid-ephemeral-runner\" is forbidden: no PriorityClass with name notexist was found"))
@@ -470,7 +470,7 @@ var _ = Describe("EphemeralRunner", func() {
 
 				var updated *v1alpha1.EphemeralRunner
 				Eventually(
-					func() (corev1.PodPhase, error) {
+					func() (v1alpha1.EphemeralRunnerPhase, error) {
 						updated = new(v1alpha1.EphemeralRunner)
 						err := k8sClient.Get(ctx, client.ObjectKey{Name: ephemeralRunner.Name, Namespace: ephemeralRunner.Namespace}, updated)
 						if err != nil {
@@ -480,7 +480,7 @@ var _ = Describe("EphemeralRunner", func() {
 					},
 					ephemeralRunnerTimeout,
 					ephemeralRunnerInterval,
-				).Should(BeEquivalentTo(phase))
+				).Should(BeEquivalentTo(v1alpha1.EphemeralRunnerPhase(phase)))
 			}
 		})
 
@@ -592,10 +592,10 @@ var _ = Describe("EphemeralRunner", func() {
 			Expect(err).To(BeNil(), "failed to patch pod status")
 
 			Consistently(
-				func() (corev1.PodPhase, error) {
+				func() (v1alpha1.EphemeralRunnerPhase, error) {
 					updated := new(v1alpha1.EphemeralRunner)
 					if err := k8sClient.Get(ctx, client.ObjectKey{Name: ephemeralRunner.Name, Namespace: ephemeralRunner.Namespace}, updated); err != nil {
-						return corev1.PodUnknown, err
+						return "", err
 					}
 					return updated.Status.Phase, nil
 				},
@@ -772,7 +772,7 @@ var _ = Describe("EphemeralRunner", func() {
 			Expect(err).To(BeNil())
 
 			Eventually(
-				func() (corev1.PodPhase, error) {
+				func() (v1alpha1.EphemeralRunnerPhase, error) {
 					updated := new(v1alpha1.EphemeralRunner)
 					if err := k8sClient.Get(ctx, client.ObjectKey{Name: ephemeralRunner.Name, Namespace: ephemeralRunner.Namespace}, updated); err != nil {
 						return "", err
@@ -781,7 +781,7 @@ var _ = Describe("EphemeralRunner", func() {
 				},
 				ephemeralRunnerTimeout,
 				ephemeralRunnerInterval,
-			).Should(BeEquivalentTo(corev1.PodRunning))
+			).Should(BeEquivalentTo(v1alpha1.EphemeralRunnerRunning))
 
 			// set phase to succeeded
 			pod.Status.Phase = corev1.PodSucceeded
@@ -789,7 +789,7 @@ var _ = Describe("EphemeralRunner", func() {
 			Expect(err).To(BeNil())
 
 			Consistently(
-				func() (corev1.PodPhase, error) {
+				func() (v1alpha1.EphemeralRunnerPhase, error) {
 					updated := new(v1alpha1.EphemeralRunner)
 					if err := k8sClient.Get(ctx, client.ObjectKey{Name: ephemeralRunner.Name, Namespace: ephemeralRunner.Namespace}, updated); err != nil {
 						return "", err
@@ -797,7 +797,7 @@ var _ = Describe("EphemeralRunner", func() {
 					return updated.Status.Phase, nil
 				},
 				ephemeralRunnerTimeout,
-			).Should(BeEquivalentTo(corev1.PodRunning))
+			).Should(BeEquivalentTo(v1alpha1.EphemeralRunnerRunning))
 		})
 	})
 

controllers/actions.github.com/ephemeralrunnerset_controller.go

@@ -162,7 +162,7 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
 		"pending", len(ephemeralRunnerState.pending),
 		"running", len(ephemeralRunnerState.running),
 		"finished", len(ephemeralRunnerState.finished),
-		"failed", len(ephemeralRunnerState.failed),
+		"failed", len(ephemeralRunnerState.aborted),
 		"deleting", len(ephemeralRunnerState.deleting),
 	)
 
@@ -185,7 +185,7 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
 			},
 			len(ephemeralRunnerState.pending),
 			len(ephemeralRunnerState.running),
-			len(ephemeralRunnerState.failed),
+			len(ephemeralRunnerState.aborted),
 		)
 	}
 
@@ -232,7 +232,7 @@ func (r *EphemeralRunnerSetReconciler) Reconcile(ctx context.Context, req ctrl.R
 		CurrentReplicas:         total,
 		PendingEphemeralRunners: len(ephemeralRunnerState.pending),
 		RunningEphemeralRunners: len(ephemeralRunnerState.running),
-		FailedEphemeralRunners:  len(ephemeralRunnerState.failed),
+		FailedEphemeralRunners:  len(ephemeralRunnerState.aborted),
 	}
 
 	// Update the status if needed.
@@ -307,13 +307,13 @@ func (r *EphemeralRunnerSetReconciler) cleanUpEphemeralRunners(ctx context.Conte
 		"pending", len(ephemeralRunnerState.pending),
 		"running", len(ephemeralRunnerState.running),
 		"finished", len(ephemeralRunnerState.finished),
-		"failed", len(ephemeralRunnerState.failed),
+		"failed", len(ephemeralRunnerState.aborted),
 		"deleting", len(ephemeralRunnerState.deleting),
 	)
 
 	log.Info("Cleanup finished or failed ephemeral runners")
 	var errs []error
-	for _, ephemeralRunner := range append(ephemeralRunnerState.finished, ephemeralRunnerState.failed...) {
+	for _, ephemeralRunner := range append(ephemeralRunnerState.finished, ephemeralRunnerState.aborted...) {
 		log.Info("Deleting ephemeral runner", "name", ephemeralRunner.Name)
 		if err := r.Delete(ctx, ephemeralRunner); err != nil && !kerrors.IsNotFound(err) {
 			errs = append(errs, err)
@@ -564,7 +564,7 @@ type ephemeralRunnerState struct {
 	pending  []*v1alpha1.EphemeralRunner
 	running  []*v1alpha1.EphemeralRunner
 	finished []*v1alpha1.EphemeralRunner
-	failed   []*v1alpha1.EphemeralRunner
+	aborted  []*v1alpha1.EphemeralRunner
 	deleting []*v1alpha1.EphemeralRunner
 
 	latestPatchID int
@@ -585,12 +585,12 @@ func newEphemeralRunnerState(ephemeralRunnerList *v1alpha1.EphemeralRunnerList)
 		}
 
 		switch r.Status.Phase {
-		case corev1.PodRunning:
+		case v1alpha1.EphemeralRunnerRunning:
 			ephemeralRunnerState.running = append(ephemeralRunnerState.running, r)
-		case corev1.PodSucceeded:
+		case v1alpha1.EphemeralRunnerSucceeded, v1alpha1.EphemeralRunnerFailed:
 			ephemeralRunnerState.finished = append(ephemeralRunnerState.finished, r)
-		case corev1.PodFailed:
+		case v1alpha1.EphemeralRunnerAborted:
-			ephemeralRunnerState.failed = append(ephemeralRunnerState.failed, r)
+			ephemeralRunnerState.aborted = append(ephemeralRunnerState.aborted, r)
 		default:
 			// Pending or no phase should be considered as pending.
 			//
@@ -603,5 +603,5 @@ func newEphemeralRunnerState(ephemeralRunnerList *v1alpha1.EphemeralRunnerList)
 }
 
 func (s *ephemeralRunnerState) scaleTotal() int {
-	return len(s.pending) + len(s.running) + len(s.failed)
+	return len(s.pending) + len(s.running) + len(s.aborted)
 }

controllers/actions.github.com/ephemeralrunnerset_controller_test.go

@@ -163,7 +163,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 					for i, runner := range runnerList.Items {
 						if runner.Status.RunnerId == 0 {
 							updatedRunner := runner.DeepCopy()
-							updatedRunner.Status.Phase = corev1.PodRunning
+							updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 							updatedRunner.Status.RunnerId = i + 100
 							err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runner))
 							Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
@@ -225,7 +225,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 					for i, runner := range runnerList.Items {
 						if runner.Status.RunnerId == 0 {
 							updatedRunner := runner.DeepCopy()
-							updatedRunner.Status.Phase = corev1.PodRunning
+							updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 							updatedRunner.Status.RunnerId = i + 100
 							err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runner))
 							Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
@@ -390,12 +390,12 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 			).Should(BeEquivalentTo(2), "2 EphemeralRunner should be created")
 
 			updatedRunner := runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodSucceeded
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerSucceeded
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
 			updatedRunner = runnerList.Items[1].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodRunning
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[1]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
@@ -468,12 +468,12 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 			).Should(BeEquivalentTo(2), "2 EphemeralRunner should be created")
 
 			updatedRunner := runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodSucceeded
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
 			updatedRunner = runnerList.Items[1].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodPending
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerPending
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[1]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
@@ -520,12 +520,12 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 			).Should(BeEquivalentTo(2), "2 EphemeralRunner should be created")
 
 			updatedRunner := runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodSucceeded
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerSucceeded
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
 			updatedRunner = runnerList.Items[1].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodRunning
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[1]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
@@ -555,7 +555,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 					}
 
 					for _, runner := range runnerList.Items {
-						if runner.Status.Phase == corev1.PodSucceeded {
+						if runner.Status.Phase == v1alpha1.EphemeralRunnerSucceeded {
 							return fmt.Errorf("Runner %s is in Succeeded phase", runner.Name)
 						}
 					}
@@ -594,12 +594,12 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 			).Should(BeEquivalentTo(2), "2 EphemeralRunner should be created")
 
 			updatedRunner := runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodSucceeded
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerSucceeded
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
 			updatedRunner = runnerList.Items[1].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodPending
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerPending
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[1]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
@@ -615,9 +615,9 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 					succeeded := 0
 					for _, runner := range runnerList.Items {
 						switch runner.Status.Phase {
-						case corev1.PodSucceeded:
+						case v1alpha1.EphemeralRunnerSucceeded:
 							succeeded++
-						case corev1.PodPending:
+						case v1alpha1.EphemeralRunnerPending:
 							pending++
 						}
 					}
@@ -657,7 +657,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 						return fmt.Errorf("Expected 1 runner, got %d", len(runnerList.Items))
 					}
 
-					if runnerList.Items[0].Status.Phase != corev1.PodPending {
+					if runnerList.Items[0].Status.Phase != v1alpha1.EphemeralRunnerPending {
 						return fmt.Errorf("Expected runner to be in Pending, got %s", runnerList.Items[0].Status.Phase)
 					}
 
@@ -669,7 +669,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 			// Now, the ephemeral runner finally is done and we can scale down to 0
 			updatedRunner = runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodSucceeded
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerSucceeded
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
@@ -715,12 +715,12 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 			// Put one runner in Pending and one in Running
 			updatedRunner := runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodPending
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerPending
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
 			updatedRunner = runnerList.Items[1].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodRunning
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[1]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
@@ -738,9 +738,9 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 					for _, runner := range runnerList.Items {
 						switch runner.Status.Phase {
-						case corev1.PodPending:
+						case v1alpha1.EphemeralRunnerPending:
 							pending++
-						case corev1.PodRunning:
+						case v1alpha1.EphemeralRunnerRunning:
 							running++
 
 						}
@@ -839,12 +839,12 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 			// Put one runner in Succeeded and one in Running
 			updatedRunner := runnerList.Items[0].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodSucceeded
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerSucceeded
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[0]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
 			updatedRunner = runnerList.Items[1].DeepCopy()
-			updatedRunner.Status.Phase = corev1.PodRunning
+			updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 			err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runnerList.Items[1]))
 			Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
 
@@ -863,9 +863,9 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 					for _, runner := range runnerList.Items {
 						switch runner.Status.Phase {
-						case corev1.PodSucceeded:
+						case v1alpha1.EphemeralRunnerSucceeded:
 							succeeded++
-						case corev1.PodRunning:
+						case v1alpha1.EphemeralRunnerRunning:
 							running++
 						}
 					}
@@ -907,7 +907,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 					}
 
 					for _, runner := range runnerList.Items {
-						if runner.Status.Phase == corev1.PodSucceeded {
+						if runner.Status.Phase == v1alpha1.EphemeralRunnerSucceeded {
 							return fmt.Errorf("Expected no runners in Succeeded phase, got one")
 						}
 					}
@@ -972,7 +972,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 						pending := pendingOriginal.DeepCopy()
 						pending.Status.RunnerId = 101
-						pending.Status.Phase = corev1.PodPending
+						pending.Status.Phase = v1alpha1.EphemeralRunnerPending
 
 						err = k8sClient.Status().Patch(ctx, pending, client.MergeFrom(pendingOriginal))
 						if err != nil {
@@ -986,7 +986,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 						empty = empty[1:]
 						running := runningOriginal.DeepCopy()
 						running.Status.RunnerId = 102
-						running.Status.Phase = corev1.PodRunning
+						running.Status.Phase = v1alpha1.EphemeralRunnerRunning
 
 						err = k8sClient.Status().Patch(ctx, running, client.MergeFrom(runningOriginal))
 						if err != nil {
@@ -1000,7 +1000,7 @@ var _ = Describe("Test EphemeralRunnerSet controller", func() {
 
 						failed := pendingOriginal.DeepCopy()
 						failed.Status.RunnerId = 103
-						failed.Status.Phase = corev1.PodFailed
+						failed.Status.Phase = v1alpha1.EphemeralRunnerAborted
 
 						err = k8sClient.Status().Patch(ctx, failed, client.MergeFrom(failedOriginal))
 						if err != nil {
@@ -1236,7 +1236,7 @@ var _ = Describe("Test EphemeralRunnerSet controller with proxy settings", func(
 				for i, runner := range runnerList.Items {
 					if runner.Status.RunnerId == 0 {
 						updatedRunner := runner.DeepCopy()
-						updatedRunner.Status.Phase = corev1.PodSucceeded
+						updatedRunner.Status.Phase = v1alpha1.EphemeralRunnerSucceeded
 						updatedRunner.Status.RunnerId = i + 100
 						err = k8sClient.Status().Patch(ctx, updatedRunner, client.MergeFrom(&runner))
 						Expect(err).NotTo(HaveOccurred(), "failed to update EphemeralRunner")
@@ -1355,7 +1355,7 @@ var _ = Describe("Test EphemeralRunnerSet controller with proxy settings", func(
 		).Should(BeEquivalentTo(1), "failed to create ephemeral runner")
 
 		runner := runnerList.Items[0].DeepCopy()
-		runner.Status.Phase = corev1.PodRunning
+		runner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 		runner.Status.RunnerId = 100
 		err = k8sClient.Status().Patch(ctx, runner, client.MergeFrom(&runnerList.Items[0]))
 		Expect(err).NotTo(HaveOccurred(), "failed to update ephemeral runner status")
@@ -1505,7 +1505,7 @@ var _ = Describe("Test EphemeralRunnerSet controller with custom root CA", func(
 		Expect(runner.Spec.GitHubServerTLS).NotTo(BeNil(), "runner tls config should not be nil")
 		Expect(runner.Spec.GitHubServerTLS).To(BeEquivalentTo(ephemeralRunnerSet.Spec.EphemeralRunnerSpec.GitHubServerTLS), "runner tls config should be correct")
 
-		runner.Status.Phase = corev1.PodRunning
+		runner.Status.Phase = v1alpha1.EphemeralRunnerRunning
 		runner.Status.RunnerId = 100
 		err = k8sClient.Status().Patch(ctx, runner, client.MergeFrom(&runnerList.Items[0]))
 		Expect(err).NotTo(HaveOccurred(), "failed to update ephemeral runner status")

controllers/actions.github.com/resourcebuilder.go

@@ -690,28 +690,20 @@ func scaleSetListenerConfigName(autoscalingListener *v1alpha1.AutoscalingListene
 	return fmt.Sprintf("%s-config", autoscalingListener.Name)
 }
 
-func hashSuffix(namespace, runnerGroup, configURL string) string {
+func scaleSetListenerName(autoscalingRunnerSet *v1alpha1.AutoscalingRunnerSet) string {
-	namespaceHash := hash.FNVHashString(namespace + "@" + runnerGroup + "@" + configURL)
+	namespaceHash := hash.FNVHashString(autoscalingRunnerSet.Namespace)
 	if len(namespaceHash) > 8 {
 		namespaceHash = namespaceHash[:8]
 	}
-	return namespaceHash
+	return fmt.Sprintf("%v-%v-listener", autoscalingRunnerSet.Name, namespaceHash)
-}
-
-func scaleSetListenerName(autoscalingRunnerSet *v1alpha1.AutoscalingRunnerSet) string {
-	return fmt.Sprintf(
-		"%v-%v-listener",
-		autoscalingRunnerSet.Name,
-		hashSuffix(
-			autoscalingRunnerSet.Namespace,
-			autoscalingRunnerSet.Spec.RunnerGroup,
-			autoscalingRunnerSet.Spec.GitHubConfigUrl,
-		),
-	)
 }
 
 func proxyListenerSecretName(autoscalingListener *v1alpha1.AutoscalingListener) string {
-	return autoscalingListener.Name + "-proxy"
+	namespaceHash := hash.FNVHashString(autoscalingListener.Spec.AutoscalingRunnerSetNamespace)
+	if len(namespaceHash) > 8 {
+		namespaceHash = namespaceHash[:8]
+	}
+	return fmt.Sprintf("%v-%v-listener-proxy", autoscalingListener.Spec.AutoscalingRunnerSetName, namespaceHash)
 }
 
 func proxyEphemeralRunnerSetSecretName(ephemeralRunnerSet *v1alpha1.EphemeralRunnerSet) string {

runner/Makefile

@@ -8,7 +8,7 @@ TARGETPLATFORM ?= $(shell arch)
 
 RUNNER_VERSION ?= 2.329.0
 RUNNER_CONTAINER_HOOKS_VERSION ?= 0.8.0
-DOCKER_VERSION ?= 28.0.4
+DOCKER_VERSION ?= 24.0.7
 
 # default list of platforms for which multiarch image is built
 ifeq (${PLATFORMS}, )

runner/actions-runner-dind-rootless.ubuntu-20.04.dockerfile

@@ -5,7 +5,7 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ENV CHANNEL=stable
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 
 # Other arguments

runner/actions-runner-dind-rootless.ubuntu-22.04.dockerfile

@@ -5,7 +5,7 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ENV CHANNEL=stable
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 ARG RUNNER_USER_UID=1001
 

runner/actions-runner-dind-rootless.ubuntu-24.04.dockerfile

@@ -5,7 +5,7 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ENV CHANNEL=stable
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 ARG RUNNER_USER_UID=1001
 

runner/actions-runner-dind.ubuntu-20.04.dockerfile

@@ -5,8 +5,8 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ARG CHANNEL=stable
-ARG DOCKER_VERSION=28.0.4
+ARG DOCKER_VERSION=24.0.7
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 
 # Use 1001 and 121 for compatibility with GitHub-hosted runners

runner/actions-runner-dind.ubuntu-22.04.dockerfile

@@ -5,8 +5,8 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ARG CHANNEL=stable
-ARG DOCKER_VERSION=28.0.4
+ARG DOCKER_VERSION=24.0.7
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 ARG RUNNER_USER_UID=1001
 ARG DOCKER_GROUP_GID=121

runner/actions-runner-dind.ubuntu-24.04.dockerfile

@@ -5,8 +5,8 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ARG CHANNEL=stable
-ARG DOCKER_VERSION=28.0.4
+ARG DOCKER_VERSION=24.0.7
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 ARG RUNNER_USER_UID=1001
 ARG DOCKER_GROUP_GID=121

runner/actions-runner.ubuntu-20.04.dockerfile

@@ -5,8 +5,8 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ARG CHANNEL=stable
-ARG DOCKER_VERSION=28.0.4
+ARG DOCKER_VERSION=24.0.7
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 
 # Use 1001 and 121 for compatibility with GitHub-hosted runners

runner/actions-runner.ubuntu-22.04.dockerfile

@@ -5,8 +5,8 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ARG CHANNEL=stable
-ARG DOCKER_VERSION=28.0.4
+ARG DOCKER_VERSION=24.0.7
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 ARG RUNNER_USER_UID=1001
 ARG DOCKER_GROUP_GID=121

runner/actions-runner.ubuntu-24.04.dockerfile

@@ -5,8 +5,8 @@ ARG RUNNER_VERSION
 ARG RUNNER_CONTAINER_HOOKS_VERSION
 # Docker and Docker Compose arguments
 ARG CHANNEL=stable
-ARG DOCKER_VERSION=28.0.4
+ARG DOCKER_VERSION=24.0.7
-ARG DOCKER_COMPOSE_VERSION=v2.38.2
+ARG DOCKER_COMPOSE_VERSION=v2.23.0
 ARG DUMB_INIT_VERSION=1.2.5
 ARG RUNNER_USER_UID=1001
 ARG DOCKER_GROUP_GID=121

test/e2e/e2e_test.go

@@ -455,7 +455,7 @@ func buildVars(repo, ubuntuVer string) vars {
 		runnerRootlessDindImage     = testing.Img(runnerRootlessDindImageRepo, runnerImageTag)
 
 		dindSidecarImageRepo = "docker"
-		dindSidecarImageTag  = "28.0.4-dind"
+		dindSidecarImageTag  = "24.0.7-dind"
 		dindSidecarImage     = testing.Img(dindSidecarImageRepo, dindSidecarImageTag)
 	)