Feature: Make Retry Period configurable

Author: viragvoros

cmd/machine-controller-manager/app/controllermanager.go

@@ -37,6 +37,7 @@ import (
 	mcmcontroller "github.com/gardener/machine-controller-manager/pkg/controller"
 	corecontroller "github.com/gardener/machine-controller-manager/pkg/util/clientbuilder/core"
 	machinecontroller "github.com/gardener/machine-controller-manager/pkg/util/clientbuilder/machine"
+	"github.com/gardener/machine-controller-manager/pkg/util/provider/machineutils"
 	kubernetesinformers "k8s.io/client-go/informers"
 	kubescheme "k8s.io/client-go/kubernetes/scheme"
 
@@ -218,6 +219,14 @@ func StartControllers(s *options.MCMServer,
 	recorder record.EventRecorder,
 	stop <-chan struct{}) error {
 
+	if s.LongRetryOverride != "" {
+		d, err := time.ParseDuration(s.LongRetryOverride)
+		if err != nil {
+			return fmt.Errorf("invalid --long-retry %q: %w", s.LongRetryOverride, err)
+		}
+		machineutils.LongRetry = machineutils.RetryPeriod(d)
+	}
+	klog.V(4).Infof("Configured LongRetry=%s", time.Duration(machineutils.LongRetry))
 	klog.V(4).Info("Getting available resources")
 	availableResources, err := getAvailableResources(controlCoreClientBuilder)
 	if err != nil {

cmd/machine-controller-manager/app/options/options.go

@@ -46,6 +46,7 @@ type MCMServer struct {
 
 	ControlKubeconfig string
 	TargetKubeconfig  string
+	LongRetryOverride string
 }
 
 // NewMCMServer creates a new MCMServer with a default config.
@@ -101,6 +102,8 @@ func (s *MCMServer) AddFlags(fs *pflag.FlagSet) {
 
 	fs.BoolVar(&s.AutoscalerScaleDownAnnotationDuringRollout, "autoscaler-scaledown-annotation-during-rollout", true, "Add cluster autoscaler scale-down disabled annotation during roll-out.")
 
+	fs.StringVar(&s.LongRetryOverride, "long-retry", "", "Override machineutils.LongRetry.")
+
 	logs.AddFlags(fs) // Here `logs` is `k8s.io/component-base/logs`.
 
 	leaderelectionconfig.BindFlags(&s.LeaderElection, fs)

pkg/util/provider/machineutils/utils.go

@@ -95,9 +95,11 @@ const (
 	// MediumRetry tells the controller to retry after a medium duration - 3 minutes
 	MediumRetry RetryPeriod = RetryPeriod(3 * time.Minute)
 	// LongRetry tells the controller to retry after a long duration - 10 minutes
-	LongRetry RetryPeriod = RetryPeriod(10 * time.Minute)
+	DefaultLongRetry RetryPeriod = RetryPeriod(10 * time.Minute)
 )
 
+var LongRetry RetryPeriod = DefaultLongRetry
+
 // EssentialTaints are taints on node object which if added/removed, require an immediate reconcile by machine controller
 // TODO: update this when taints for ALT updation and PostCreate operations is introduced.
 var EssentialTaints = []string{TaintNodeCriticalComponentsNotReady}