✨ add 10% jitter to ResyncPeriod

pkg/cache/cache.go

@@ -93,7 +93,10 @@ type Options struct {
 	// Mapper is the RESTMapper to use for mapping GroupVersionKinds to Resources
 	Mapper meta.RESTMapper
 
-	// Resync is the resync period. Defaults to defaultResyncTime.
+	// Resync is the base frequency the informers are resynced.
+	// Defaults to defaultResyncTime.
+	// A 10 percent jitter will be added to the Resync period between informers
+	// So that all informers will not send list requests simultaneously.
 	Resync *time.Duration
 
 	// Namespace restricts the cache's ListWatch to the desired namespace

pkg/cache/internal/informers_map.go

@@ -18,6 +18,7 @@ package internal
 
 import (
 	"fmt"
+	"math/rand"
 	"sync"
 	"time"
 
@@ -92,7 +93,9 @@ type specificInformersMap struct {
 	// stop is the stop channel to stop informers
 	stop <-chan struct{}
 
-	// resync is the frequency the informers are resynced
+	// resync is the base frequency the informers are resynced
+	// a 10 percent jitter will be added to the resync period between informers
+	// so that all informers will not send list requests simultaneously.
 	resync time.Duration
 
 	// mu guards access to the map
@@ -188,7 +191,7 @@ func (ip *specificInformersMap) addInformerToMap(gvk schema.GroupVersionKind, ob
 	if err != nil {
 		return nil, false, err
 	}
-	ni := cache.NewSharedIndexInformer(lw, obj, ip.resync, cache.Indexers{
+	ni := cache.NewSharedIndexInformer(lw, obj, resyncPeriod(ip.resync)(), cache.Indexers{
 		cache.NamespaceIndex: cache.MetaNamespaceIndexFunc,
 	})
 	i := &MapEntry{
@@ -274,3 +277,14 @@ func createUnstructuredListWatch(gvk schema.GroupVersionKind, ip *specificInform
 		},
 	}, nil
 }
+
+// resyncPeriod returns a function which generates a duration each time it is
+// invoked; this is so that multiple controllers don't get into lock-step and all
+// hammer the apiserver with list requests simultaneously.
+func resyncPeriod(resync time.Duration) func() time.Duration {
+	return func() time.Duration {
+		// the factor will fall into [0.9, 1.1)
+		factor := rand.Float64()/5.0 + 0.9
+		return time.Duration(float64(resync.Nanoseconds()) * factor)
+	}
+}

pkg/manager/manager.go

@@ -111,6 +111,8 @@ type Options struct {
 	// reconciled. A lower period will correct entropy more quickly, but reduce
 	// responsiveness to change if there are many watched resources. Change this
 	// value only if you know what you are doing. Defaults to 10 hours if unset.
+	// there will a 10 percent jitter between the SyncPeriod of all controllers
+	// so that all controllers will not send list requests simultaneously.
 	SyncPeriod *time.Duration
 
 	// LeaderElection determines whether or not to use leader election when