Expose backoff time parameters

cluster-autoscaler/clusterstate/clusterstate.go

@@ -42,15 +42,6 @@ import (
 const (
 	// MaxNodeStartupTime is the maximum time from the moment the node is registered to the time the node is ready.
 	MaxNodeStartupTime = 15 * time.Minute
-
-	// MaxNodeGroupBackoffDuration is the maximum backoff duration for a NodeGroup after new nodes failed to start.
-	MaxNodeGroupBackoffDuration = 30 * time.Minute
-
-	// InitialNodeGroupBackoffDuration is the duration of first backoff after a new node failed to start.
-	InitialNodeGroupBackoffDuration = 5 * time.Minute
-
-	// NodeGroupBackoffResetTimeout is the time after last failed scale-up when the backoff duration is reset.
-	NodeGroupBackoffResetTimeout = 3 * time.Hour
 )
 
 // ScaleUpRequest contains information about the requested node group scale up.

cluster-autoscaler/clusterstate/clusterstate_test.go

@@ -653,7 +653,7 @@ func TestUpdateLastTransitionTimes(t *testing.T) {
 		}
 	}
 
-	expectedNgTimestamps := make(map[string](map[api.ClusterAutoscalerConditionType]metav1.Time), 0)
+	expectedNgTimestamps := make(map[string]map[api.ClusterAutoscalerConditionType]metav1.Time, 0)
 	// Same as cluster-wide
 	expectedNgTimestamps["ng1"] = map[api.ClusterAutoscalerConditionType]metav1.Time{
 		api.ClusterAutoscalerHealth:    now,
@@ -710,7 +710,7 @@ func TestScaleUpBackoff(t *testing.T) {
 	assert.False(t, clusterstate.IsNodeGroupSafeToScaleUp(ng1, now))
 
 	// Backoff should expire after timeout
-	now = now.Add(InitialNodeGroupBackoffDuration).Add(time.Second)
+	now = now.Add(5 * time.Minute /*InitialNodeGroupBackoffDuration*/).Add(time.Second)
 	assert.True(t, clusterstate.IsClusterHealthy())
 	assert.True(t, clusterstate.IsNodeGroupHealthy("ng1"))
 	assert.True(t, clusterstate.IsNodeGroupSafeToScaleUp(ng1, now))
@@ -724,7 +724,7 @@ func TestScaleUpBackoff(t *testing.T) {
 	assert.True(t, clusterstate.IsNodeGroupHealthy("ng1"))
 	assert.False(t, clusterstate.IsNodeGroupSafeToScaleUp(ng1, now))
 
-	now = now.Add(InitialNodeGroupBackoffDuration).Add(time.Second)
+	now = now.Add(5 * time.Minute /*InitialNodeGroupBackoffDuration*/).Add(time.Second)
 	assert.False(t, clusterstate.IsNodeGroupSafeToScaleUp(ng1, now))
 
 	// The backoff should be cleared after a successful scale-up
@@ -873,5 +873,6 @@ func TestScaleUpFailures(t *testing.T) {
 }
 
 func newBackoff() backoff.Backoff {
-	return backoff.NewIdBasedExponentialBackoff(InitialNodeGroupBackoffDuration, MaxNodeGroupBackoffDuration, NodeGroupBackoffResetTimeout)
+	return backoff.NewIdBasedExponentialBackoff(5*time.Minute, /*InitialNodeGroupBackoffDuration*/
+		30*time.Minute /*MaxNodeGroupBackoffDuration*/, 3*time.Hour /*NodeGroupBackoffResetTimeout*/)
 }

cluster-autoscaler/config/autoscaling_options.go

@@ -171,4 +171,10 @@ type AutoscalingOptions struct {
 	DaemonSetEvictionForOccupiedNodes bool
 	// User agent to use for HTTP calls.
 	UserAgent string
+	// InitialNodeGroupBackoffDuration is the duration of first backoff after a new node failed to start
+	InitialNodeGroupBackoffDuration time.Duration
+	// MaxNodeGroupBackoffDuration is the maximum backoff duration for a NodeGroup after new nodes failed to start.
+	MaxNodeGroupBackoffDuration time.Duration
+	// NodeGroupBackoffResetTimeout is the time after last failed scale-up when the backoff duration is reset.
+	NodeGroupBackoffResetTimeout time.Duration
 }

cluster-autoscaler/core/autoscaler.go

@@ -22,7 +22,6 @@ import (
 
 	"k8s.io/autoscaler/cluster-autoscaler/cloudprovider"
 	cloudBuilder "k8s.io/autoscaler/cluster-autoscaler/cloudprovider/builder"
-	"k8s.io/autoscaler/cluster-autoscaler/clusterstate"
 	"k8s.io/autoscaler/cluster-autoscaler/config"
 	"k8s.io/autoscaler/cluster-autoscaler/context"
 	"k8s.io/autoscaler/cluster-autoscaler/debuggingsnapshot"
@@ -121,7 +120,7 @@ func initializeDefaultOptions(opts *AutoscalerOptions) error {
 	}
 	if opts.Backoff == nil {
 		opts.Backoff =
-			backoff.NewIdBasedExponentialBackoff(clusterstate.InitialNodeGroupBackoffDuration, clusterstate.MaxNodeGroupBackoffDuration, clusterstate.NodeGroupBackoffResetTimeout)
+			backoff.NewIdBasedExponentialBackoff(opts.InitialNodeGroupBackoffDuration, opts.MaxNodeGroupBackoffDuration, opts.NodeGroupBackoffResetTimeout)
 	}
 
 	return nil

cluster-autoscaler/core/test/common.go

@@ -20,6 +20,7 @@ import (
 	"fmt"
 	"reflect"
 	"testing"
+	"time"
 
 	"k8s.io/autoscaler/cluster-autoscaler/debuggingsnapshot"
 
@@ -53,7 +54,6 @@ import (
 	"github.com/stretchr/testify/assert"
 
 	apiv1 "k8s.io/api/core/v1"
-	"k8s.io/autoscaler/cluster-autoscaler/clusterstate"
 	"k8s.io/autoscaler/cluster-autoscaler/utils/backoff"
 	kube_client "k8s.io/client-go/kubernetes"
 	kube_record "k8s.io/client-go/tools/record"
@@ -296,5 +296,6 @@ func (p *MockAutoprovisioningNodeGroupListProcessor) CleanUp() {
 
 // NewBackoff creates a new backoff object
 func NewBackoff() backoff.Backoff {
-	return backoff.NewIdBasedExponentialBackoff(clusterstate.InitialNodeGroupBackoffDuration, clusterstate.MaxNodeGroupBackoffDuration, clusterstate.NodeGroupBackoffResetTimeout)
+	return backoff.NewIdBasedExponentialBackoff(5*time.Minute, /*InitialNodeGroupBackoffDuration*/
+		30*time.Minute /*MaxNodeGroupBackoffDuration*/, 3*time.Hour /*NodeGroupBackoffResetTimeout*/)
 }

cluster-autoscaler/main.go

@@ -62,7 +62,7 @@ import (
 	componentbaseconfig "k8s.io/component-base/config"
 	"k8s.io/component-base/config/options"
 	"k8s.io/component-base/metrics/legacyregistry"
-	klog "k8s.io/klog/v2"
+	"k8s.io/klog/v2"
 )
 
 // MultiStringFlag is a flag for passing multiple parameters using same flag
@@ -188,10 +188,16 @@ var (
 	daemonSetEvictionForEmptyNodes     = flag.Bool("daemonset-eviction-for-empty-nodes", false, "DaemonSet pods will be gracefully terminated from empty nodes")
 	daemonSetEvictionForOccupiedNodes  = flag.Bool("daemonset-eviction-for-occupied-nodes", true, "DaemonSet pods will be gracefully terminated from non-empty nodes")
 	userAgent                          = flag.String("user-agent", "cluster-autoscaler", "User agent used for HTTP calls.")
+	emitPerNodeGroupMetrics            = flag.Bool("emit-per-nodegroup-metrics", false, "If true, emit per node group metrics.")
+	debuggingSnapshotEnabled           = flag.Bool("debugging-snapshot-enabled", false, "Whether the debugging snapshot of cluster autoscaler feature is enabled")
+	nodeInfoCacheExpireTime            = flag.Duration("node-info-cache-expire-time", 87600*time.Hour, "Node Info cache expire time for each item. Default value is 10 years.")
 
-	emitPerNodeGroupMetrics  = flag.Bool("emit-per-nodegroup-metrics", false, "If true, emit per node group metrics.")
+	initialNodeGroupBackoffDuration = flag.Duration("initial-node-group-backoff-duration", 5*time.Minute,
-	debuggingSnapshotEnabled = flag.Bool("debugging-snapshot-enabled", false, "Whether the debugging snapshot of cluster autoscaler feature is enabled")
+		"initialNodeGroupBackoffDuration is the duration of first backoff after a new node failed to start.")
-	nodeInfoCacheExpireTime  = flag.Duration("node-info-cache-expire-time", 87600*time.Hour, "Node Info cache expire time for each item. Default value is 10 years.")
+	maxNodeGroupBackoffDuration = flag.Duration("max-node-group-backoff-duration", 30*time.Minute,
+		"maxNodeGroupBackoffDuration is the maximum backoff duration for a NodeGroup after new nodes failed to start.")
+	nodeGroupBackoffResetTimeout = flag.Duration("node-group-backoff-reset-timeout", 3*time.Hour,
+		"nodeGroupBackoffResetTimeout is the time after last failed scale-up when the backoff duration is reset.")
 )
 
 func createAutoscalingOptions() config.AutoscalingOptions {
@@ -272,6 +278,9 @@ func createAutoscalingOptions() config.AutoscalingOptions {
 		DaemonSetEvictionForEmptyNodes:     *daemonSetEvictionForEmptyNodes,
 		DaemonSetEvictionForOccupiedNodes:  *daemonSetEvictionForOccupiedNodes,
 		UserAgent:                          *userAgent,
+		InitialNodeGroupBackoffDuration:    *initialNodeGroupBackoffDuration,
+		MaxNodeGroupBackoffDuration:        *maxNodeGroupBackoffDuration,
+		NodeGroupBackoffResetTimeout:       *nodeGroupBackoffResetTimeout,
 	}
 }