making CrashesToMarkUnhealthy optional

cmd/e2e/build_crashing_test.go

@@ -43,16 +43,12 @@ var _ = Describe("Crashing Build", func() {
 			g.Expect(verifyGameServerBuildOverall(ctx, kubeClient, state)).To(Succeed())
 		}, 45*time.Second, interval).Should(Succeed()) // bigger timeout because of the time crashes take to occur and captured by the controller
 
-		// we are updating the GameServerBuild with a big CrashesToMarkUnhealthy to give time to the GameServerBuild to become Healthy
-		// Reasoning: At one point during running the e2e tests, we noticed that this test failed.
-		// This is because the GameServers crashed multiple (10) times so the GameServerBuild stayed Unhealthy,
-		// before having the chance to transition (temporarily) to Healthy. So, by setting it to 1000 we give it more chance to transition to Healthy,
-		// before decreasing it to 10 (in the next step) so that it can become Unhealthy again.
+		// we are updating the GameServerBuild with a nil CrashesToMarkUnhealthy so the GameServerBuild will become Healthy
 		gsb := &mpsv1alpha1.GameServerBuild{}
 		err = kubeClient.Get(ctx, client.ObjectKey{Name: testBuildCrashingName, Namespace: testNamespace}, gsb)
 		Expect(err).ToNot(HaveOccurred())
 		patch := client.MergeFrom(gsb.DeepCopy())
-		gsb.Spec.CrashesToMarkUnhealthy = 1000
+		gsb.Spec.CrashesToMarkUnhealthy = nil
 		err = kubeClient.Patch(ctx, gsb, patch)
 		Expect(err).ToNot(HaveOccurred())
 
@@ -64,12 +60,13 @@ var _ = Describe("Crashing Build", func() {
 			g.Expect(gsb.Status.Health).To(Equal(mpsv1alpha1.BuildHealthy))
 		}, 10*time.Second, interval).Should(Succeed())
 
-		// we're decreasing the CrashesToMarkUnhealthy to 10 so that the
+		// we're setting the CrashesToMarkUnhealthy to 10 so that the
 		// GameServerBuild will eventually become Unhealthy
 		err = kubeClient.Get(ctx, client.ObjectKey{Name: testBuildCrashingName, Namespace: testNamespace}, gsb)
 		Expect(err).ToNot(HaveOccurred())
 		patch = client.MergeFrom(gsb.DeepCopy())
-		gsb.Spec.CrashesToMarkUnhealthy = 10
+		crashes := 10
+		gsb.Spec.CrashesToMarkUnhealthy = &crashes
 		err = kubeClient.Patch(ctx, gsb, patch)
 		Expect(err).ToNot(HaveOccurred())
 
@@ -97,6 +94,7 @@ var _ = Describe("Crashing Build", func() {
 func createCrashingBuild(buildName, buildID, img string) *mpsv1alpha1.GameServerBuild {
 	gsb := createTestBuild(buildName, buildID, img)
 	gsb.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "sleep 2 && command_that_does_not_exist"}
-	gsb.Spec.CrashesToMarkUnhealthy = 5
+	crashes := 5
+	gsb.Spec.CrashesToMarkUnhealthy = &crashes
 	return gsb
 }

cmd/e2e/build_unhealthy_gameservers_test.go

@@ -308,7 +308,7 @@ var _ = Describe("GameServerBuild with GameServers without Gsdk", func() {
 func createBuildWithoutGsdk(buildName, buildID, img string) *mpsv1alpha1.GameServerBuild {
 	gsb := createTestBuild(buildName, buildID, img)
 	gsb.Spec.Template.Spec.Containers[0].Command = []string{"/bin/sh", "-c", "sleep 3600"}
-	gsb.Spec.CrashesToMarkUnhealthy = 2
-
+	crashes := 2
+	gsb.Spec.CrashesToMarkUnhealthy = &crashes
 	return gsb
 }

installfiles/operator.yaml

@@ -81,7 +81,6 @@ spec:
                   type: object
                 type: array
               crashesToMarkUnhealthy:
-                default: 5
                 description: CrashesToMarkUnhealthy is the number of crashes needed to mark the build unhealthy
                 minimum: 0
                 type: integer

installfiles/operator_with_monitoring.yaml

@@ -81,7 +81,6 @@ spec:
                   type: object
                 type: array
               crashesToMarkUnhealthy:
-                default: 5
                 description: CrashesToMarkUnhealthy is the number of crashes needed to mark the build unhealthy
                 minimum: 0
                 type: integer

installfiles/operator_with_security.yaml

@@ -81,7 +81,6 @@ spec:
                   type: object
                 type: array
               crashesToMarkUnhealthy:
-                default: 5
                 description: CrashesToMarkUnhealthy is the number of crashes needed to mark the build unhealthy
                 minimum: 0
                 type: integer

installfiles/operator_with_security_and_monitoring.yaml

@@ -81,7 +81,6 @@ spec:
                   type: object
                 type: array
               crashesToMarkUnhealthy:
-                default: 5
                 description: CrashesToMarkUnhealthy is the number of crashes needed to mark the build unhealthy
                 minimum: 0
                 type: integer

pkg/operator/api/v1alpha1/gameserverbuild_types.go

@@ -67,10 +67,9 @@ type GameServerBuildSpec struct {
 	// PortsToExpose is an array of ports that will be exposed on the VM
 	PortsToExpose []int32 `json:"portsToExpose"`
 
-	//+kubebuilder:default=5
 	//+kubebuilder:validation:Minimum=0
 	// CrashesToMarkUnhealthy is the number of crashes needed to mark the build unhealthy
-	CrashesToMarkUnhealthy int `json:"crashesToMarkUnhealthy,omitempty"`
+	CrashesToMarkUnhealthy *int `json:"crashesToMarkUnhealthy,omitempty"`
 
 	// BuildMetadata is the metadata for this GameServerBuild
 	BuildMetadata []BuildMetadataItem `json:"buildMetadata,omitempty"`

pkg/operator/config/crd/bases/mps.playfab.com_gameserverbuilds.yaml

@@ -69,7 +69,6 @@ spec:
                   type: object
                 type: array
               crashesToMarkUnhealthy:
-                default: 5
                 description: CrashesToMarkUnhealthy is the number of crashes needed
                   to mark the build unhealthy
                 minimum: 0

pkg/operator/controllers/gameserverbuild_controller.go

@@ -106,7 +106,9 @@ func (r *GameServerBuildReconciler) Reconcile(ctx context.Context, req ctrl.Requ
 	}
 
 	// if GameServerBuild is unhealthy and current crashes equal or more than CrashesToMarkUnhealthy, do nothing more
-	if gsb.Status.Health == mpsv1alpha1.BuildUnhealthy && gsb.Status.CrashesCount >= gsb.Spec.CrashesToMarkUnhealthy {
+	if gsb.Status.Health == mpsv1alpha1.BuildUnhealthy &&
+		gsb.Spec.CrashesToMarkUnhealthy != nil &&
+		gsb.Status.CrashesCount >= *gsb.Spec.CrashesToMarkUnhealthy {
 		log.Info("GameServerBuild is Unhealthy, do nothing")
 		r.Recorder.Event(&gsb, corev1.EventTypeNormal, "Unhealthy Build", "GameServerBuild is Unhealthy, stopping reconciliation")
 		return ctrl.Result{}, nil
@@ -256,7 +258,8 @@ func (r *GameServerBuildReconciler) updateStatus(ctx context.Context, gsb *mpsv1
 		gsb.Status.CrashesCount = existingCrashes + crashesCount
 		gsb.Status.CurrentStandingByReadyDesired = fmt.Sprintf("%d/%d", standingByCount, gsb.Spec.StandingBy)
 
-		if gsb.Status.CrashesCount >= gsb.Spec.CrashesToMarkUnhealthy {
+		// GameServerBuild can only be set as Unhealthy if CrashesToMarkUnhealthy has been explicitly been set by the user
+		if gsb.Spec.CrashesToMarkUnhealthy != nil && gsb.Status.CrashesCount >= *gsb.Spec.CrashesToMarkUnhealthy {
 			gsb.Status.Health = mpsv1alpha1.BuildUnhealthy
 		} else {
 			gsb.Status.Health = mpsv1alpha1.BuildHealthy

pkg/operator/controllers/gameserverbuild_controller_test.go

@@ -153,6 +153,8 @@ var _ = Describe("GameServerBuild controller tests", func() {
 			// create a Build with 6 standingBy
 			buildName, buildID := getNewBuildNameAndID()
 			gsb := testGenerateGameServerBuild(buildName, testnamespace, buildID, 6, 6, false)
+			crashes := 5
+			gsb.Spec.CrashesToMarkUnhealthy = &crashes
 			Expect(testk8sClient.Create(ctx, &gsb)).Should(Succeed())
 			testVerifyTotalGameServerCount(ctx, buildID, 6)
 			testUpdateInitializingGameServersToStandingBy(ctx, buildID)