add liveness probe and adjust cpu/mem limit (#1530)

Author: M00nF1sh

config/controller/controller.yaml

@@ -29,10 +29,18 @@ spec:
             - --ingress-class=alb
           resources:
             limits:
-              cpu: 100m
-              memory: 2000Mi
+              cpu: 200m
+              memory: 500Mi
             requests:
               cpu: 100m
               memory: 200Mi
+          livenessProbe:
+            failureThreshold: 2
+            httpGet:
+              path: /healthz
+              port: 61779
+              scheme: HTTP
+            initialDelaySeconds: 30
+            timeoutSeconds: 10
       terminationGracePeriodSeconds: 10
       serviceAccountName: controller

main.go

@@ -39,9 +39,10 @@ import (
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/runtime"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/targetgroupbinding"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/version"
-	ctrl "sigs.k8s.io/controller-runtime"
 	corewebhook "sigs.k8s.io/aws-load-balancer-controller/webhooks/core"
 	elbv2webhook "sigs.k8s.io/aws-load-balancer-controller/webhooks/elbv2"
+	ctrl "sigs.k8s.io/controller-runtime"
+	"sigs.k8s.io/controller-runtime/pkg/healthz"
 	"sigs.k8s.io/controller-runtime/pkg/log/zap"
 	"sigs.k8s.io/controller-runtime/pkg/metrics"
 	// +kubebuilder:scaffold:imports
@@ -126,6 +127,14 @@ func main() {
 		os.Exit(1)
 	}
 
+	// Add liveness probe
+	err = mgr.AddHealthzCheck("health-ping", healthz.Ping)
+	setupLog.Info("adding health check for controller")
+	if err != nil {
+		setupLog.Error(err, "unable add a health check")
+		os.Exit(1)
+	}
+
 	podReadinessGateInjector := inject.NewPodReadinessGate(controllerCFG.PodWebhookConfig,
 		mgr.GetClient(), ctrl.Log.WithName("pod-readiness-gate-injector"))
 	corewebhook.NewPodMutator(podReadinessGateInjector).SetupWithManager(mgr)

pkg/config/runtime_config.go

@@ -12,6 +12,7 @@ import (
 
 const (
 	flagMetricsBindAddr         = "metrics-bind-addr"
+	flagHealthProbeBindAddr     = "health-probe-bind-addr"
 	flagWebhookBindPort         = "webhook-bind-port"
 	flagEnableLeaderElection    = "enable-leader-election"
 	flagLeaderElectionID        = "leader-election-id"
@@ -25,6 +26,7 @@ const (
 	defaultLeaderElectionNamespace = ""
 	defaultWatchNamespace          = corev1.NamespaceAll
 	defaultMetricsAddr             = ":8080"
+	defaultHealthProbeBindAddress  = ":61779"
 	defaultSyncPeriod              = 60 * time.Minute
 	defaultWebhookBindPort         = 9443
 	// High enough QPS to fit all expected use cases. QPS=0 is not set here, because
@@ -41,6 +43,7 @@ type RuntimeConfig struct {
 	KubeConfig              string
 	WebhookBindPort         int
 	MetricsBindAddress      string
+	HealthProbeBindAddress  string
 	EnableLeaderElection    bool
 	LeaderElectionID        string
 	LeaderElectionNamespace string
@@ -54,6 +57,8 @@ func (c *RuntimeConfig) BindFlags(fs *pflag.FlagSet) {
 		"Path to the kubeconfig file containing authorization and API server information.")
 	fs.StringVar(&c.MetricsBindAddress, flagMetricsBindAddr, defaultMetricsAddr,
 		"The address the metric endpoint binds to.")
+	fs.StringVar(&c.HealthProbeBindAddress, flagHealthProbeBindAddr, defaultHealthProbeBindAddress,
+		"The address the health probes binds to.")
 	fs.IntVar(&c.WebhookBindPort, flagWebhookBindPort, defaultWebhookBindPort,
 		"The TCP port the Webhook server binds to.")
 	fs.BoolVar(&c.EnableLeaderElection, flagEnableLeaderElection, true,
@@ -94,6 +99,7 @@ func BuildRuntimeOptions(rtCfg RuntimeConfig, scheme *runtime.Scheme) ctrl.Optio
 		Scheme:                  scheme,
 		Port:                    rtCfg.WebhookBindPort,
 		MetricsBindAddress:      rtCfg.MetricsBindAddress,
+		HealthProbeBindAddress:  rtCfg.HealthProbeBindAddress,
 		LeaderElection:          rtCfg.EnableLeaderElection,
 		LeaderElectionID:        rtCfg.LeaderElectionID,
 		LeaderElectionNamespace: rtCfg.LeaderElectionNamespace,